diff --git a/configs/config_1.14G_dp128_tp2_pp1_acc2_mbs2_seq2048_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp128_tp2_pp1_acc2_mbs2_seq2048_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..358e68e34f783280a303b26a53be330959caa4f0 --- /dev/null +++ b/configs/config_1.14G_dp128_tp2_pp1_acc2_mbs2_seq2048_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp128_tp2_pp1_acc2_mbs2_seq2048_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 128 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp128_tp2_pp1_acc4_mbs4_seq2048_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp128_tp2_pp1_acc4_mbs4_seq2048_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..62a13b57f088426c6debb8bf9e06c846500e982e --- /dev/null +++ b/configs/config_1.14G_dp128_tp2_pp1_acc4_mbs4_seq2048_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp128_tp2_pp1_acc4_mbs4_seq2048_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 128 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp16_tp1_pp1_acc4_mbs8_seq2048_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp16_tp1_pp1_acc4_mbs8_seq2048_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..cb2901e91c1823c0b728dc6ee4cb16492eb81a9a --- /dev/null +++ b/configs/config_1.14G_dp16_tp1_pp1_acc4_mbs8_seq2048_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp16_tp1_pp1_acc4_mbs8_seq2048_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 16 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp16_tp8_pp1_acc8_mbs16_seq2048_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp16_tp8_pp1_acc8_mbs16_seq2048_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0818fb02fce88d7cce189b7c61d9a6b44bcdecd2 --- /dev/null +++ b/configs/config_1.14G_dp16_tp8_pp1_acc8_mbs16_seq2048_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp16_tp8_pp1_acc8_mbs16_seq2048_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 16 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp2_tp256_pp1_acc1_mbs256_seq2048_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp2_tp256_pp1_acc1_mbs256_seq2048_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f32cc4f4b43f079803539c6754b9ebdbb241500e --- /dev/null +++ b/configs/config_1.14G_dp2_tp256_pp1_acc1_mbs256_seq2048_zero1_tpmodeALL_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp2_tp256_pp1_acc1_mbs256_seq2048_zero1_tpmodeALL_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 256 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 256 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp2_tp256_pp1_acc256_mbs1_seq2048_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp2_tp256_pp1_acc256_mbs1_seq2048_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..39bcb3a15b7aae900d4e607fe0bea7b65bc6084f --- /dev/null +++ b/configs/config_1.14G_dp2_tp256_pp1_acc256_mbs1_seq2048_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp2_tp256_pp1_acc256_mbs1_seq2048_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 256 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 256 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp2_tp256_pp1_acc4_mbs4_seq32768_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp2_tp256_pp1_acc4_mbs4_seq32768_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..695f70e9a0c27594eb9dcc18d226f3548060f48a --- /dev/null +++ b/configs/config_1.14G_dp2_tp256_pp1_acc4_mbs4_seq32768_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp2_tp256_pp1_acc4_mbs4_seq32768_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 256 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp2_tp4_pp1_acc8_mbs128_seq2048_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp2_tp4_pp1_acc8_mbs128_seq2048_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c41f6ac6e0361cab87f5f80cc052cb178345b9af --- /dev/null +++ b/configs/config_1.14G_dp2_tp4_pp1_acc8_mbs128_seq2048_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp2_tp4_pp1_acc8_mbs128_seq2048_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 128 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp2_tp8_pp1_acc128_mbs2_seq2048_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp2_tp8_pp1_acc128_mbs2_seq2048_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..31f48daeb4c630b202bb5c6ebf2f4d16f54587b4 --- /dev/null +++ b/configs/config_1.14G_dp2_tp8_pp1_acc128_mbs2_seq2048_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp2_tp8_pp1_acc128_mbs2_seq2048_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 128 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp4_tp16_pp1_acc64_mbs8_seq2048_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp4_tp16_pp1_acc64_mbs8_seq2048_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f5a05421d48d23e7d9681a3521941b6fcbce3b30 --- /dev/null +++ b/configs/config_1.14G_dp4_tp16_pp1_acc64_mbs8_seq2048_zero1_tpmodeALL_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp4_tp16_pp1_acc64_mbs8_seq2048_zero1_tpmodeALL_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 64 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp4_tp1_pp2_acc1_mbs32_seq32768_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp4_tp1_pp2_acc1_mbs32_seq32768_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f336b7932c72c8fb7c4c9d29d5a4e9b6f47580b5 --- /dev/null +++ b/configs/config_1.14G_dp4_tp1_pp2_acc1_mbs32_seq32768_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp4_tp1_pp2_acc1_mbs32_seq32768_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 32 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp4_tp32_pp1_acc16_mbs2_seq32768_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp4_tp32_pp1_acc16_mbs2_seq32768_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c7c8fe39abd94e27cba5c446ac32b51c54ee5813 --- /dev/null +++ b/configs/config_1.14G_dp4_tp32_pp1_acc16_mbs2_seq32768_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp4_tp32_pp1_acc16_mbs2_seq32768_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 32 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp4_tp64_pp1_acc32_mbs1_seq32768_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp4_tp64_pp1_acc32_mbs1_seq32768_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..42769361b65389340d4427846a0a838738fd8b26 --- /dev/null +++ b/configs/config_1.14G_dp4_tp64_pp1_acc32_mbs1_seq32768_zero1_tpmodeALL_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp4_tp64_pp1_acc32_mbs1_seq32768_zero1_tpmodeALL_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 64 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp4_tp8_pp1_acc8_mbs16_seq8192_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp4_tp8_pp1_acc8_mbs16_seq8192_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5e4bad2e2010c2b7a92b31a1d6ff3b6dec9d1d4c --- /dev/null +++ b/configs/config_1.14G_dp4_tp8_pp1_acc8_mbs16_seq8192_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp4_tp8_pp1_acc8_mbs16_seq8192_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp512_tp1_pp1_acc1_mbs1_seq8192_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp512_tp1_pp1_acc1_mbs1_seq8192_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8f2294313445ab41869947d725a5cde862d362b2 --- /dev/null +++ b/configs/config_1.14G_dp512_tp1_pp1_acc1_mbs1_seq8192_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp512_tp1_pp1_acc1_mbs1_seq8192_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 512 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp64_tp1_pp1_acc2_mbs4_seq8192_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp64_tp1_pp1_acc2_mbs4_seq8192_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..392fb3df8eb0a9a9da990a966a3d6e83c96b4226 --- /dev/null +++ b/configs/config_1.14G_dp64_tp1_pp1_acc2_mbs4_seq8192_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp64_tp1_pp1_acc2_mbs4_seq8192_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 64 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp64_tp1_pp1_acc4_mbs2_seq8192_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp64_tp1_pp1_acc4_mbs2_seq8192_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c041fb306019a59527d2a85dba381c1508cc19e2 --- /dev/null +++ b/configs/config_1.14G_dp64_tp1_pp1_acc4_mbs2_seq8192_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp64_tp1_pp1_acc4_mbs2_seq8192_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 64 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp8_tp1_pp1_acc1_mbs16_seq8192_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp8_tp1_pp1_acc1_mbs16_seq8192_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d2039ee26cf19140d54495afee0e75396438b102 --- /dev/null +++ b/configs/config_1.14G_dp8_tp1_pp1_acc1_mbs16_seq8192_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp8_tp1_pp1_acc1_mbs16_seq8192_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp128_tp4_pp1_acc1_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp128_tp4_pp1_acc1_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d579834dce841f0e16b6be173c3d98f31471f4f9 --- /dev/null +++ b/configs/config_1.34G_dp128_tp4_pp1_acc1_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp128_tp4_pp1_acc1_mbs2_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 128 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp128_tp4_pp1_acc4_mbs1_seq2048_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp128_tp4_pp1_acc4_mbs1_seq2048_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..97fb1234256e90b7c5429c8db9337d65cf23ae5f --- /dev/null +++ b/configs/config_1.34G_dp128_tp4_pp1_acc4_mbs1_seq2048_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp128_tp4_pp1_acc4_mbs1_seq2048_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 128 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp16_tp2_pp1_acc8_mbs1_seq32768_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp16_tp2_pp1_acc8_mbs1_seq32768_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8c3260f219a48d74fe7430a487d4ed49706d7fc7 --- /dev/null +++ b/configs/config_1.34G_dp16_tp2_pp1_acc8_mbs1_seq32768_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp16_tp2_pp1_acc8_mbs1_seq32768_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 16 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp16_tp32_pp1_acc16_mbs2_seq2048_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp16_tp32_pp1_acc16_mbs2_seq2048_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d4f0d2e72bafa9a2c91e952ecc90c51770efe006 --- /dev/null +++ b/configs/config_1.34G_dp16_tp32_pp1_acc16_mbs2_seq2048_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp16_tp32_pp1_acc16_mbs2_seq2048_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 16 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 32 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp1_tp1_pp8_acc16_mbs16_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp1_tp1_pp8_acc16_mbs16_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..687cac054518796c4e792becb5aab655b9b6d093 --- /dev/null +++ b/configs/config_1.34G_dp1_tp1_pp8_acc16_mbs16_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp1_tp1_pp8_acc16_mbs16_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 8 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp1_tp4_pp2_acc4_mbs64_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp1_tp4_pp2_acc4_mbs64_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c4ad140a62dcf12a546e69ba0f7445ca21e088b7 --- /dev/null +++ b/configs/config_1.34G_dp1_tp4_pp2_acc4_mbs64_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp1_tp4_pp2_acc4_mbs64_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 64 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp2_tp16_pp1_acc32_mbs2_seq32768_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp2_tp16_pp1_acc32_mbs2_seq32768_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b053029b0f4fbdf345b30849b4891d9cb42a3295 --- /dev/null +++ b/configs/config_1.34G_dp2_tp16_pp1_acc32_mbs2_seq32768_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp2_tp16_pp1_acc32_mbs2_seq32768_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp2_tp1_pp4_acc16_mbs1_seq32768_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp2_tp1_pp4_acc16_mbs1_seq32768_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e34a1091b930342deeb9747cd94a646d2e105ebc --- /dev/null +++ b/configs/config_1.34G_dp2_tp1_pp4_acc16_mbs1_seq32768_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: bench_seqlen.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp2_tp1_pp4_acc16_mbs1_seq32768_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp2_tp1_pp4_acc16_mbs64_seq2048_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp2_tp1_pp4_acc16_mbs64_seq2048_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4eca95c537016eebc285f809cb9128bf99b0e88f --- /dev/null +++ b/configs/config_1.34G_dp2_tp1_pp4_acc16_mbs64_seq2048_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: bench_seqlen.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp2_tp1_pp4_acc16_mbs64_seq2048_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 64 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp2_tp8_pp1_acc128_mbs2_seq8192_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp2_tp8_pp1_acc128_mbs2_seq8192_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5314e19b884f1bb0ada7e4a2919accf225298d85 --- /dev/null +++ b/configs/config_1.34G_dp2_tp8_pp1_acc128_mbs2_seq8192_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp2_tp8_pp1_acc128_mbs2_seq8192_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 128 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp2_tp8_pp1_acc4_mbs16_seq32768_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp2_tp8_pp1_acc4_mbs16_seq32768_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..96e9131ec559c0d7c31b2ff675d8b05b4be31a0c --- /dev/null +++ b/configs/config_1.34G_dp2_tp8_pp1_acc4_mbs16_seq32768_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp2_tp8_pp1_acc4_mbs16_seq32768_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp32_tp1_pp1_acc8_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp32_tp1_pp1_acc8_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..394b7a34c77a1ccb566d2ec960b2d082a212f3e7 --- /dev/null +++ b/configs/config_1.34G_dp32_tp1_pp1_acc8_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp32_tp1_pp1_acc8_mbs1_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 32 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp32_tp1_pp4_acc4_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp32_tp1_pp4_acc4_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..260df9c585bdfe8dc4d87a862c8d50af42dda2c6 --- /dev/null +++ b/configs/config_1.34G_dp32_tp1_pp4_acc4_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp32_tp1_pp4_acc4_mbs2_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 32 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp4_tp2_pp1_acc4_mbs128_seq2048_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp4_tp2_pp1_acc4_mbs128_seq2048_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..387df6e16c795c65c4d295ba31c270f7ead22e24 --- /dev/null +++ b/configs/config_1.34G_dp4_tp2_pp1_acc4_mbs128_seq2048_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp4_tp2_pp1_acc4_mbs128_seq2048_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 128 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp4_tp2_pp1_acc64_mbs8_seq2048_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp4_tp2_pp1_acc64_mbs8_seq2048_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..64108efd619e095d92d28d9720f5a2d263fdf0e2 --- /dev/null +++ b/configs/config_1.34G_dp4_tp2_pp1_acc64_mbs8_seq2048_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp4_tp2_pp1_acc64_mbs8_seq2048_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 64 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp4_tp32_pp1_acc2_mbs256_seq2048_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp4_tp32_pp1_acc2_mbs256_seq2048_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8b06a5fc947c2f362055e69605db634c6fa8c8d0 --- /dev/null +++ b/configs/config_1.34G_dp4_tp32_pp1_acc2_mbs256_seq2048_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp4_tp32_pp1_acc2_mbs256_seq2048_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 32 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 256 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp64_tp2_pp1_acc2_mbs1_seq8192_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp64_tp2_pp1_acc2_mbs1_seq8192_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c2700732b7ab7af7475d365e6ed343704e3c9fef --- /dev/null +++ b/configs/config_1.34G_dp64_tp2_pp1_acc2_mbs1_seq8192_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp64_tp2_pp1_acc2_mbs1_seq8192_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 64 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp8_tp2_pp1_acc1_mbs64_seq2048_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp8_tp2_pp1_acc1_mbs64_seq2048_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0afdee97ad9e5e171d01040aa486add7528830c0 --- /dev/null +++ b/configs/config_1.34G_dp8_tp2_pp1_acc1_mbs64_seq2048_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp8_tp2_pp1_acc1_mbs64_seq2048_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 64 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp8_tp64_pp1_acc2_mbs8_seq8192_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp8_tp64_pp1_acc2_mbs8_seq8192_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0954e4e8c0e1b41785bbc28798f8f27ca8dd77d2 --- /dev/null +++ b/configs/config_1.34G_dp8_tp64_pp1_acc2_mbs8_seq8192_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp8_tp64_pp1_acc2_mbs8_seq8192_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 64 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp8_tp8_pp1_acc2_mbs128_seq2048_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp8_tp8_pp1_acc2_mbs128_seq2048_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..2e2bfcafdeaf4c192f9c1179555adcc3d91fe20c --- /dev/null +++ b/configs/config_1.34G_dp8_tp8_pp1_acc2_mbs128_seq2048_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp8_tp8_pp1_acc2_mbs128_seq2048_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 128 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_11.4G_dp1_tp8_pp1_acc1_mbs1_seq16384_zero0_tpmodeRED_l42_h4096_heads32.yaml b/configs/config_11.4G_dp1_tp8_pp1_acc1_mbs1_seq16384_zero0_tpmodeRED_l42_h4096_heads32.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ed5dd18d2b36d57c7aa9e6cf1391d396c2fa206f --- /dev/null +++ b/configs/config_11.4G_dp1_tp8_pp1_acc1_mbs1_seq16384_zero0_tpmodeRED_l42_h4096_heads32.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: bench_seqlen.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 11.4G_dp1_tp8_pp1_acc1_mbs1_seq16384_zero0_tpmodeRED_l42_h4096_heads32 + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 16384 + is_llama_config: true + max_position_embeddings: 16384 + num_attention_heads: 32 + num_hidden_layers: 42 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 16384 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_2.28G_dp4_tp8_pp1_acc4_mbs16_seq2048_zero1_tpmodeALL_l26_h2304_heads16.yaml b/configs/config_2.28G_dp4_tp8_pp1_acc4_mbs16_seq2048_zero1_tpmodeALL_l26_h2304_heads16.yaml new file mode 100644 index 0000000000000000000000000000000000000000..9c538e8edc09d08082ba8df38df18d269fb5e13d --- /dev/null +++ b/configs/config_2.28G_dp4_tp8_pp1_acc4_mbs16_seq2048_zero1_tpmodeALL_l26_h2304_heads16.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: bench_tp.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 2.28G_dp4_tp8_pp1_acc4_mbs16_seq2048_zero1_tpmodeALL_l26_h2304_heads16 + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2304 + initializer_range: 0.02 + intermediate_size: 9216 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 16 + num_hidden_layers: 26 + num_key_value_heads: 16 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.27G_dp64_tp8_pp1_acc1_mbs1_seq2048_zero0_tpmodeRED_l28_h3072_heads24.yaml b/configs/config_3.27G_dp64_tp8_pp1_acc1_mbs1_seq2048_zero0_tpmodeRED_l28_h3072_heads24.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e8c3c37929c355e08154091865a7a2e227f659d6 --- /dev/null +++ b/configs/config_3.27G_dp64_tp8_pp1_acc1_mbs1_seq2048_zero0_tpmodeRED_l28_h3072_heads24.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: bench_seqlen.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.27G_dp64_tp8_pp1_acc1_mbs1_seq2048_zero0_tpmodeRED_l28_h3072_heads24 + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 24 + num_hidden_layers: 28 + num_key_value_heads: 24 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 64 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.56G_dp8_tp8_pp1_acc2_mbs16_seq4096_zero1_l28_h3072_heads24.yaml b/configs/config_3.56G_dp8_tp8_pp1_acc2_mbs16_seq4096_zero1_l28_h3072_heads24.yaml new file mode 100644 index 0000000000000000000000000000000000000000..802548ca2940ee21a84e286dff42af2ee6f1c243 --- /dev/null +++ b/configs/config_3.56G_dp8_tp8_pp1_acc2_mbs16_seq4096_zero1_l28_h3072_heads24.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: bench_elie_4k.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.56G_dp8_tp8_pp1_acc2_mbs16_seq4096_zero1_l28_h3072_heads24 + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 24 + num_hidden_layers: 28 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 128256 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.57G_dp1_tp1_pp4_acc3_mbs1_seq4096_zero0_tpmodeRED_vocab131k_cache.yaml b/configs/config_3.57G_dp1_tp1_pp4_acc3_mbs1_seq4096_zero0_tpmodeRED_vocab131k_cache.yaml new file mode 100644 index 0000000000000000000000000000000000000000..738ebe63bd88a65a656790e44069d8a7a553e3bc --- /dev/null +++ b/configs/config_3.57G_dp1_tp1_pp4_acc3_mbs1_seq4096_zero0_tpmodeRED_vocab131k_cache.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.57G_dp1_tp1_pp4_acc3_mbs1_seq4096_zero0_tpmodeRED_vocab131k_cache + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 28 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 3 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.57G_dp2_tp1_pp4_acc16_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_3.57G_dp2_tp1_pp4_acc16_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..9446f521fc4110e23c3ab9480458009e1743d7ec --- /dev/null +++ b/configs/config_3.57G_dp2_tp1_pp4_acc16_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.57G_dp2_tp1_pp4_acc16_mbs8_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 28 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.57G_dp32_tp1_pp2_acc4_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_3.57G_dp32_tp1_pp2_acc4_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8f55a9bb766fcbe9dab8babb8504838982c722ca --- /dev/null +++ b/configs/config_3.57G_dp32_tp1_pp2_acc4_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.57G_dp32_tp1_pp2_acc4_mbs2_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 28 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 32 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.57G_dp32_tp4_pp2_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_3.57G_dp32_tp4_pp2_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..35ccef7cbce95762bb8f9017fea25f568f281a3c --- /dev/null +++ b/configs/config_3.57G_dp32_tp4_pp2_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.57G_dp32_tp4_pp2_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 28 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 32 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 128 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.57G_dp32_tp8_pp1_acc1_mbs8_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_3.57G_dp32_tp8_pp1_acc1_mbs8_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0d98aa4f0800bc0b7e03fa8670fbeb2395c2c02b --- /dev/null +++ b/configs/config_3.57G_dp32_tp8_pp1_acc1_mbs8_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.57G_dp32_tp8_pp1_acc1_mbs8_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 28 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 32 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.57G_dp8_tp2_pp2_acc1_mbs32_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_3.57G_dp8_tp2_pp2_acc1_mbs32_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..434cf1f845fd141475a391ac3da5eb02ca39327c --- /dev/null +++ b/configs/config_3.57G_dp8_tp2_pp2_acc1_mbs32_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.57G_dp8_tp2_pp2_acc1_mbs32_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 28 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 32 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_37.8G_dp8_tp4_pp16_acc1_mbs1_seq2048_zero0_tpmodeRED_l80_h8192_heads64.yaml b/configs/config_37.8G_dp8_tp4_pp16_acc1_mbs1_seq2048_zero0_tpmodeRED_l80_h8192_heads64.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e9c0e84793958d1f31615bf21f161f3725b11b49 --- /dev/null +++ b/configs/config_37.8G_dp8_tp4_pp16_acc1_mbs1_seq2048_zero0_tpmodeRED_l80_h8192_heads64.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: bench_seqlen.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 37.8G_dp8_tp4_pp16_acc1_mbs1_seq2048_zero0_tpmodeRED_l80_h8192_heads64 + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 8192 + initializer_range: 0.02 + intermediate_size: 131072 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 64 + num_hidden_layers: 80 + num_key_value_heads: 64 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 16 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_469G_dp16_tp16_pp1_acc4_mbs4_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_469G_dp16_tp16_pp1_acc4_mbs4_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d4bc9445c2cb013a958c3749bbe01caadd3a66bf --- /dev/null +++ b/configs/config_469G_dp16_tp16_pp1_acc4_mbs4_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 469G_dp16_tp16_pp1_acc4_mbs4_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 16384 + initializer_range: 0.02 + intermediate_size: 53248 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 128 + num_hidden_layers: 126 + num_key_value_heads: 128 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 16 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_469G_dp16_tp4_pp1_acc16_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_469G_dp16_tp4_pp1_acc16_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..566516549a3af3d3cfc3d4f1a6eb4b118c779f71 --- /dev/null +++ b/configs/config_469G_dp16_tp4_pp1_acc16_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 469G_dp16_tp4_pp1_acc16_mbs1_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 16384 + initializer_range: 0.02 + intermediate_size: 53248 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 128 + num_hidden_layers: 126 + num_key_value_heads: 128 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 16 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_469G_dp1_tp32_pp4_acc4_mbs64_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_469G_dp1_tp32_pp4_acc4_mbs64_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0e9ec89a0ba19b73ce750564499b38eb090d6fcc --- /dev/null +++ b/configs/config_469G_dp1_tp32_pp4_acc4_mbs64_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 469G_dp1_tp32_pp4_acc4_mbs64_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 16384 + initializer_range: 0.02 + intermediate_size: 53248 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 128 + num_hidden_layers: 126 + num_key_value_heads: 128 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 32 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 64 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_469G_dp4_tp16_pp2_acc4_mbs16_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_469G_dp4_tp16_pp2_acc4_mbs16_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..14afe48c8eefb4bd9c67c670024368329f147860 --- /dev/null +++ b/configs/config_469G_dp4_tp16_pp2_acc4_mbs16_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 469G_dp4_tp16_pp2_acc4_mbs16_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 16384 + initializer_range: 0.02 + intermediate_size: 53248 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 128 + num_hidden_layers: 126 + num_key_value_heads: 128 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_469G_dp4_tp1_pp4_acc16_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_469G_dp4_tp1_pp4_acc16_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..525268d8cb5d5eeafaf007356f32b8681acf02e0 --- /dev/null +++ b/configs/config_469G_dp4_tp1_pp4_acc16_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 469G_dp4_tp1_pp4_acc16_mbs4_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 16384 + initializer_range: 0.02 + intermediate_size: 53248 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 128 + num_hidden_layers: 126 + num_key_value_heads: 128 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_469G_dp8_tp1_pp8_acc32_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_469G_dp8_tp1_pp8_acc32_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b7b2a6ca1597213b63b2829ab4dbb50bceb1cbf2 --- /dev/null +++ b/configs/config_469G_dp8_tp1_pp8_acc32_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 469G_dp8_tp1_pp8_acc32_mbs1_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 16384 + initializer_range: 0.02 + intermediate_size: 53248 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 128 + num_hidden_layers: 126 + num_key_value_heads: 128 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 8 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp1_tp16_pp4_acc1_mbs256_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp1_tp16_pp4_acc1_mbs256_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d07faea93dd4a3ea463833486e5cb43a5149eacb --- /dev/null +++ b/configs/config_8.86G_dp1_tp16_pp4_acc1_mbs256_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp1_tp16_pp4_acc1_mbs256_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 256 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp1_tp2_pp8_acc16_mbs16_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp1_tp2_pp8_acc16_mbs16_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c010773c5bf07b6e4a84d261f5c9319bd9375eeb --- /dev/null +++ b/configs/config_8.86G_dp1_tp2_pp8_acc16_mbs16_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp1_tp2_pp8_acc16_mbs16_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 8 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp1_tp8_pp1_acc1_mbs8_seq4096_zero0_tpmodeRED_vocab131k_cache.yaml b/configs/config_8.86G_dp1_tp8_pp1_acc1_mbs8_seq4096_zero0_tpmodeRED_vocab131k_cache.yaml new file mode 100644 index 0000000000000000000000000000000000000000..42350a99b4b20a9301a3906221c167436151ae80 --- /dev/null +++ b/configs/config_8.86G_dp1_tp8_pp1_acc1_mbs8_seq4096_zero0_tpmodeRED_vocab131k_cache.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp1_tp8_pp1_acc1_mbs8_seq4096_zero0_tpmodeRED_vocab131k_cache + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp256_tp1_pp2_acc1_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp256_tp1_pp2_acc1_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a584a74823829a47de5f9f365d8deed2ac1c1165 --- /dev/null +++ b/configs/config_8.86G_dp256_tp1_pp2_acc1_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp256_tp1_pp2_acc1_mbs1_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 256 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp2_tp1_pp16_acc128_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp2_tp1_pp16_acc128_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..fd8530b4bc030375cac33bc032409c473086bdb5 --- /dev/null +++ b/configs/config_8.86G_dp2_tp1_pp16_acc128_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp2_tp1_pp16_acc128_mbs1_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 16 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 128 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp32_tp1_pp1_acc1_mbs8_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp32_tp1_pp1_acc1_mbs8_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..2f68ecc0e947a1cbc7413cae43411ac749d7469b --- /dev/null +++ b/configs/config_8.86G_dp32_tp1_pp1_acc1_mbs8_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp32_tp1_pp1_acc1_mbs8_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 32 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp32_tp8_pp2_acc4_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp32_tp8_pp2_acc4_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e8b93db7de4a54fabf07eb55d610a57578e24600 --- /dev/null +++ b/configs/config_8.86G_dp32_tp8_pp2_acc4_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp32_tp8_pp2_acc4_mbs2_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 32 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp64_tp4_pp2_acc2_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp64_tp4_pp2_acc2_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8a8519aefbb2ad2d575defd74ab5b69a0844308c --- /dev/null +++ b/configs/config_8.86G_dp64_tp4_pp2_acc2_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp64_tp4_pp2_acc2_mbs2_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 64 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_80G_dp16_tp2_pp1_acc16_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp16_tp2_pp1_acc16_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..bbc71591cc7fffbab38d992bd06e3b6b73c0e48e --- /dev/null +++ b/configs/config_80G_dp16_tp2_pp1_acc16_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 80G_dp16_tp2_pp1_acc16_mbs1_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 8192 + initializer_range: 0.02 + intermediate_size: 28672 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 64 + num_hidden_layers: 80 + num_key_value_heads: 64 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 16 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_80G_dp16_tp4_pp4_acc8_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp16_tp4_pp4_acc8_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..dd4b0db5c39b65f232f7245289cbb1505f67ed51 --- /dev/null +++ b/configs/config_80G_dp16_tp4_pp4_acc8_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 80G_dp16_tp4_pp4_acc8_mbs2_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 8192 + initializer_range: 0.02 + intermediate_size: 28672 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 64 + num_hidden_layers: 80 + num_key_value_heads: 64 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 16 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_80G_dp1_tp2_pp4_acc32_mbs8_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp1_tp2_pp4_acc32_mbs8_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0f771b963d52e1fbf0828ff6e747cd11684f0333 --- /dev/null +++ b/configs/config_80G_dp1_tp2_pp4_acc32_mbs8_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 80G_dp1_tp2_pp4_acc32_mbs8_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 8192 + initializer_range: 0.02 + intermediate_size: 28672 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 64 + num_hidden_layers: 80 + num_key_value_heads: 64 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_80G_dp2_tp2_pp16_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp2_tp2_pp16_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f86893a86b4ad95c05ef31f168f7f3107bf1f243 --- /dev/null +++ b/configs/config_80G_dp2_tp2_pp16_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 80G_dp2_tp2_pp16_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 8192 + initializer_range: 0.02 + intermediate_size: 28672 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 64 + num_hidden_layers: 80 + num_key_value_heads: 64 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 16 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 128 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_80G_dp2_tp4_pp1_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp2_tp4_pp1_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..7de6028fb970cbeddc548d68bc401b45b54085be --- /dev/null +++ b/configs/config_80G_dp2_tp4_pp1_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 80G_dp2_tp4_pp1_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 8192 + initializer_range: 0.02 + intermediate_size: 28672 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 64 + num_hidden_layers: 80 + num_key_value_heads: 64 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 128 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_80G_dp2_tp4_pp2_acc128_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp2_tp4_pp2_acc128_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..31ebb6ec63052a2c5eab0b2b85e1e16925d5cc79 --- /dev/null +++ b/configs/config_80G_dp2_tp4_pp2_acc128_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 80G_dp2_tp4_pp2_acc128_mbs1_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 8192 + initializer_range: 0.02 + intermediate_size: 28672 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 64 + num_hidden_layers: 80 + num_key_value_heads: 64 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 128 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_80G_dp2_tp8_pp16_acc16_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp2_tp8_pp16_acc16_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..efdd6a5769322e1f8aaa9522b4b37a369182d86e --- /dev/null +++ b/configs/config_80G_dp2_tp8_pp16_acc16_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 80G_dp2_tp8_pp16_acc16_mbs8_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 8192 + initializer_range: 0.02 + intermediate_size: 28672 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 64 + num_hidden_layers: 80 + num_key_value_heads: 64 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 16 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_80G_dp8_tp2_pp2_acc2_mbs16_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp8_tp2_pp2_acc2_mbs16_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0a43ae5b9c0be70314e885c4accc373f0d474f92 --- /dev/null +++ b/configs/config_80G_dp8_tp2_pp2_acc2_mbs16_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 80G_dp8_tp2_pp2_acc2_mbs16_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 8192 + initializer_range: 0.02 + intermediate_size: 28672 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 64 + num_hidden_layers: 80 + num_key_value_heads: 64 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_80G_dp8_tp8_pp4_acc8_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp8_tp8_pp4_acc8_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1f6d8a5a0d72d0404446b7b2320f0ee776694999 --- /dev/null +++ b/configs/config_80G_dp8_tp8_pp4_acc8_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 80G_dp8_tp8_pp4_acc8_mbs4_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 8192 + initializer_range: 0.02 + intermediate_size: 28672 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 64 + num_hidden_layers: 80 + num_key_value_heads: 64 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_dp16_tp8_pp1_acc1_mbs1_seq65536.yaml b/configs/config_dp16_tp8_pp1_acc1_mbs1_seq65536.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d8ca83f681189a12de32c3ffd7b4227bccc80ee7 --- /dev/null +++ b/configs/config_dp16_tp8_pp1_acc1_mbs1_seq65536.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: bench.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: dp16_tp8_pp1_acc1_mbs1_seq65536 + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 65536 + num_attention_heads: 32 + num_hidden_layers: 24 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 49152 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 16 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 65536 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_dp4_tp8_pp1_acc1_mbs1_seq65536.yaml b/configs/config_dp4_tp8_pp1_acc1_mbs1_seq65536.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f8bcec5e5269d38979fcf3bdd753d95b5c2ae51c --- /dev/null +++ b/configs/config_dp4_tp8_pp1_acc1_mbs1_seq65536.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: bench.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: dp4_tp8_pp1_acc1_mbs1_seq65536 + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 65536 + num_attention_heads: 32 + num_hidden_layers: 24 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 49152 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 65536 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_dp512_tp1_pp1_acc1_mbs2_seq4096_zero0_l24_h2048_heads16.yaml b/configs/config_dp512_tp1_pp1_acc1_mbs2_seq4096_zero0_l24_h2048_heads16.yaml new file mode 100644 index 0000000000000000000000000000000000000000..7035a04b1e50f86a9f152a3c454b246219bfbaf6 --- /dev/null +++ b/configs/config_dp512_tp1_pp1_acc1_mbs2_seq4096_zero0_l24_h2048_heads16.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: bench.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: dp512_tp1_pp1_acc1_mbs2_seq4096_zero0_l24_h2048_heads16 + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 16 + num_hidden_layers: 24 + num_key_value_heads: 16 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 49152 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 512 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100