large_crafting_sft_fail

This model is a fine-tuned version of meta-llama/Meta-Llama-3-8B-Instruct on the identity and the large_crafting_sft_fail datasets. It achieves the following results on the evaluation set:

• Loss: 0.3223

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 1
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 16
• total_train_batch_size: 128
• total_eval_batch_size: 16
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1.0

Training results

Training Loss	Epoch	Step	Validation Loss
0.5429	0.0323	50	0.4980
0.5398	0.0646	100	0.4740
0.5484	0.0969	150	0.4833
0.5265	0.1291	200	0.4780
0.5278	0.1614	250	0.4793
0.5259	0.1937	300	0.4519
0.5293	0.2260	350	0.4497
0.5098	0.2583	400	0.4303
0.482	0.2906	450	0.4249
0.4683	0.3229	500	0.4224
0.4572	0.3552	550	0.4136
0.456	0.3874	600	0.4034
0.4606	0.4197	650	0.3983
0.4285	0.4520	700	0.3874
0.4499	0.4843	750	0.3806
0.4198	0.5166	800	0.3685
0.4208	0.5489	850	0.3661
0.4379	0.5812	900	0.3637
0.4075	0.6134	950	0.3558
0.4121	0.6457	1000	0.3513
0.4112	0.6780	1050	0.3454
0.4041	0.7103	1100	0.3457
0.3852	0.7426	1150	0.3384
0.3656	0.7749	1200	0.3340
0.384	0.8072	1250	0.3303
0.3605	0.8395	1300	0.3276
0.3593	0.8717	1350	0.3247
0.3624	0.9040	1400	0.3233
0.3734	0.9363	1450	0.3229
0.3609	0.9686	1500	0.3223

Framework versions

• Transformers 4.49.0
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0