Add new CrossEncoder model

README.md

@@ -0,0 +1,396 @@
+---
+language:
+- en
+license: apache-2.0
+tags:
+- sentence-transformers
+- cross-encoder
+- text-classification
+- generated_from_trainer
+- dataset_size:578402
+- loss:BinaryCrossEntropyLoss
+pipeline_tag: text-classification
+library_name: sentence-transformers
+metrics:
+- map
+- mrr@10
+- ndcg@10
+model-index:
+- name: NeoBERT-medium trained on GooAQ
+  results: []
+---
+
+# NeoBERT-medium trained on GooAQ
+
+This is a [Cross Encoder](https://www.sbert.net/docs/cross_encoder/usage/usage.html) model trained using the [sentence-transformers](https://www.SBERT.net) library. It computes scores for pairs of texts, which can be used for semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.
+
+## Model Details
+
+### Model Description
+- **Model Type:** Cross Encoder
+<!-- - **Base model:** [Unknown](https://huggingface.co/unknown) -->
+- **Maximum Sequence Length:** 4096 tokens
+- **Number of Output Labels:** 1 label
+<!-- - **Training Dataset:** Unknown -->
+- **Language:** en
+- **License:** apache-2.0
+
+### Model Sources
+
+- **Documentation:** [Sentence Transformers Documentation](https://sbert.net)
+- **Documentation:** [Cross Encoder Documentation](https://www.sbert.net/docs/cross_encoder/usage/usage.html)
+- **Repository:** [Sentence Transformers on GitHub](https://github.com/UKPLab/sentence-transformers)
+- **Hugging Face:** [Cross Encoders on Hugging Face](https://huggingface.co/models?library=sentence-transformers&other=cross-encoder)
+
+## Usage
+
+### Direct Usage (Sentence Transformers)
+
+First install the Sentence Transformers library:
+
+```bash
+pip install -U sentence-transformers
+```
+
+Then you can load this model and run inference.
+```python
+from sentence_transformers import CrossEncoder
+
+# Download from the 🤗 Hub
+model = CrossEncoder("tomaarsen/reranker-NeoBERT-gooaq-bce")
+# Get scores for pairs of texts
+pairs = [
+    ['what are the signs of a bad yeast infection?', "['Itching and irritation in the vagina and vulva.', 'A burning sensation, especially during intercourse or while urinating.', 'Redness and swelling of the vulva.', 'Vaginal pain and soreness.', 'Vaginal rash.', 'Thick, white, odor-free vaginal discharge with a cottage cheese appearance.', 'Watery vaginal discharge.']"],
+    ['what are the signs of a bad yeast infection?', 'Vaginal yeast infections can cause: itching and irritation in the vagina. redness, swelling, or itching of the vulva (the folds of skin outside the vagina) a thick, white discharge that can look like cottage cheese and is usually odorless, although it might smell like bread or yeast.'],
+    ['what are the signs of a bad yeast infection?', 'It can feel like itching or maybe even burning. Or you may experience swelling so extreme, it leads to sores. Whether your symptoms are mild or severe, a yeast infection can be uncomfortable. Also known as vaginal candidiasis, yeast infections are caused by a fungus.'],
+    ['what are the signs of a bad yeast infection?', 'Complications of untreated yeast infections If left untreated, vaginal candidiasis will most likely get worse, causing itching, redness, and inflammation in the area surrounding your vagina. This may lead to a skin infection if the inflamed area becomes cracked, or if continual scratching creates open or raw areas.'],
+    ['what are the signs of a bad yeast infection?', "Drinking alcohol may also put you at greater risk for yeast infections. So if you're worried about yeast infection symptoms, consider curbing your cocktails. Eating only yeast-free foods is one way some women try to control yeast infections."],
+]
+scores = model.predict(pairs)
+print(scores.shape)
+# (5,)
+
+# Or rank different texts based on similarity to a single text
+ranks = model.rank(
+    'what are the signs of a bad yeast infection?',
+    [
+        "['Itching and irritation in the vagina and vulva.', 'A burning sensation, especially during intercourse or while urinating.', 'Redness and swelling of the vulva.', 'Vaginal pain and soreness.', 'Vaginal rash.', 'Thick, white, odor-free vaginal discharge with a cottage cheese appearance.', 'Watery vaginal discharge.']",
+        'Vaginal yeast infections can cause: itching and irritation in the vagina. redness, swelling, or itching of the vulva (the folds of skin outside the vagina) a thick, white discharge that can look like cottage cheese and is usually odorless, although it might smell like bread or yeast.',
+        'It can feel like itching or maybe even burning. Or you may experience swelling so extreme, it leads to sores. Whether your symptoms are mild or severe, a yeast infection can be uncomfortable. Also known as vaginal candidiasis, yeast infections are caused by a fungus.',
+        'Complications of untreated yeast infections If left untreated, vaginal candidiasis will most likely get worse, causing itching, redness, and inflammation in the area surrounding your vagina. This may lead to a skin infection if the inflamed area becomes cracked, or if continual scratching creates open or raw areas.',
+        "Drinking alcohol may also put you at greater risk for yeast infections. So if you're worried about yeast infection symptoms, consider curbing your cocktails. Eating only yeast-free foods is one way some women try to control yeast infections.",
+    ]
+)
+# [{'corpus_id': ..., 'score': ...}, {'corpus_id': ..., 'score': ...}, ...]
+```
+
+<!--
+### Direct Usage (Transformers)
+
+<details><summary>Click to see the direct usage in Transformers</summary>
+
+</details>
+-->
+
+<!--
+### Downstream Usage (Sentence Transformers)
+
+You can finetune this model on your own dataset.
+
+<details><summary>Click to expand</summary>
+
+</details>
+-->
+
+<!--
+### Out-of-Scope Use
+
+*List how the model may foreseeably be misused and address what users ought not to do with the model.*
+-->
+
+## Evaluation
+
+### Metrics
+
+#### Cross Encoder Reranking
+
+* Dataset: `gooaq-dev`
+* Evaluated with [<code>CrossEncoderRerankingEvaluator</code>](https://sbert.net/docs/package_reference/cross_encoder/evaluation.html#sentence_transformers.cross_encoder.evaluation.CrossEncoderRerankingEvaluator)
+
+| Metric      | Value                |
+|:------------|:---------------------|
+| map         | 0.8039 (+0.2728)     |
+| mrr@10      | 0.8028 (+0.2789)     |
+| **ndcg@10** | **0.8475 (+0.2562)** |
+
+<!--
+## Bias, Risks and Limitations
+
+*What are the known or foreseeable issues stemming from this model? You could also flag here known failure cases or weaknesses of the model.*
+-->
+
+<!--
+### Recommendations
+
+*What are recommendations with respect to the foreseeable issues? For example, filtering explicit content.*
+-->
+
+## Training Details
+
+### Training Dataset
+
+#### Unnamed Dataset
+
+* Size: 578,402 training samples
+* Columns: <code>question</code>, <code>answer</code>, and <code>label</code>
+* Approximate statistics based on the first 1000 samples:
+  |         | question                                                                                       | answer                                                                                          | label                                           |
+  |:--------|:-----------------------------------------------------------------------------------------------|:------------------------------------------------------------------------------------------------|:------------------------------------------------|
+  | type    | string                                                                                         | string                                                                                          | int                                             |
+  | details | <ul><li>min: 21 characters</li><li>mean: 43.81 characters</li><li>max: 91 characters</li></ul> | <ul><li>min: 51 characters</li><li>mean: 251.2 characters</li><li>max: 365 characters</li></ul> | <ul><li>0: ~82.90%</li><li>1: ~17.10%</li></ul> |
+* Samples:
+  | question                                                  | answer                                                                                                                                                                                                                                                                                                                                      | label          |
+  |:----------------------------------------------------------|:--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|:---------------|
+  | <code>what are the signs of a bad yeast infection?</code> | <code>['Itching and irritation in the vagina and vulva.', 'A burning sensation, especially during intercourse or while urinating.', 'Redness and swelling of the vulva.', 'Vaginal pain and soreness.', 'Vaginal rash.', 'Thick, white, odor-free vaginal discharge with a cottage cheese appearance.', 'Watery vaginal discharge.']</code> | <code>1</code> |
+  | <code>what are the signs of a bad yeast infection?</code> | <code>Vaginal yeast infections can cause: itching and irritation in the vagina. redness, swelling, or itching of the vulva (the folds of skin outside the vagina) a thick, white discharge that can look like cottage cheese and is usually odorless, although it might smell like bread or yeast.</code>                                   | <code>0</code> |
+  | <code>what are the signs of a bad yeast infection?</code> | <code>It can feel like itching or maybe even burning. Or you may experience swelling so extreme, it leads to sores. Whether your symptoms are mild or severe, a yeast infection can be uncomfortable. Also known as vaginal candidiasis, yeast infections are caused by a fungus.</code>                                                    | <code>0</code> |
+* Loss: [<code>BinaryCrossEntropyLoss</code>](https://sbert.net/docs/package_reference/cross_encoder/losses.html#binarycrossentropyloss) with these parameters:
+  ```json
+  {
+      "activation_fct": "torch.nn.modules.linear.Identity",
+      "pos_weight": 5
+  }
+  ```
+
+### Training Hyperparameters
+#### Non-Default Hyperparameters
+
+- `eval_strategy`: steps
+- `per_device_train_batch_size`: 64
+- `per_device_eval_batch_size`: 64
+- `learning_rate`: 2e-05
+- `num_train_epochs`: 1
+- `warmup_ratio`: 0.1
+- `seed`: 12
+- `bf16`: True
+- `dataloader_num_workers`: 4
+- `load_best_model_at_end`: True
+
+#### All Hyperparameters
+<details><summary>Click to expand</summary>
+
+- `overwrite_output_dir`: False
+- `do_predict`: False
+- `eval_strategy`: steps
+- `prediction_loss_only`: True
+- `per_device_train_batch_size`: 64
+- `per_device_eval_batch_size`: 64
+- `per_gpu_train_batch_size`: None
+- `per_gpu_eval_batch_size`: None
+- `gradient_accumulation_steps`: 1
+- `eval_accumulation_steps`: None
+- `torch_empty_cache_steps`: None
+- `learning_rate`: 2e-05
+- `weight_decay`: 0.0
+- `adam_beta1`: 0.9
+- `adam_beta2`: 0.999
+- `adam_epsilon`: 1e-08
+- `max_grad_norm`: 1.0
+- `num_train_epochs`: 1
+- `max_steps`: -1
+- `lr_scheduler_type`: linear
+- `lr_scheduler_kwargs`: {}
+- `warmup_ratio`: 0.1
+- `warmup_steps`: 0
+- `log_level`: passive
+- `log_level_replica`: warning
+- `log_on_each_node`: True
+- `logging_nan_inf_filter`: True
+- `save_safetensors`: True
+- `save_on_each_node`: False
+- `save_only_model`: False
+- `restore_callback_states_from_checkpoint`: False
+- `no_cuda`: False
+- `use_cpu`: False
+- `use_mps_device`: False
+- `seed`: 12
+- `data_seed`: None
+- `jit_mode_eval`: False
+- `use_ipex`: False
+- `bf16`: True
+- `fp16`: False
+- `fp16_opt_level`: O1
+- `half_precision_backend`: auto
+- `bf16_full_eval`: False
+- `fp16_full_eval`: False
+- `tf32`: None
+- `local_rank`: 0
+- `ddp_backend`: None
+- `tpu_num_cores`: None
+- `tpu_metrics_debug`: False
+- `debug`: []
+- `dataloader_drop_last`: False
+- `dataloader_num_workers`: 4
+- `dataloader_prefetch_factor`: None
+- `past_index`: -1
+- `disable_tqdm`: False
+- `remove_unused_columns`: True
+- `label_names`: None
+- `load_best_model_at_end`: True
+- `ignore_data_skip`: False
+- `fsdp`: []
+- `fsdp_min_num_params`: 0
+- `fsdp_config`: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
+- `fsdp_transformer_layer_cls_to_wrap`: None
+- `accelerator_config`: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
+- `deepspeed`: None
+- `label_smoothing_factor`: 0.0
+- `optim`: adamw_torch
+- `optim_args`: None
+- `adafactor`: False
+- `group_by_length`: False
+- `length_column_name`: length
+- `ddp_find_unused_parameters`: None
+- `ddp_bucket_cap_mb`: None
+- `ddp_broadcast_buffers`: False
+- `dataloader_pin_memory`: True
+- `dataloader_persistent_workers`: False
+- `skip_memory_metrics`: True
+- `use_legacy_prediction_loop`: False
+- `push_to_hub`: False
+- `resume_from_checkpoint`: None
+- `hub_model_id`: None
+- `hub_strategy`: every_save
+- `hub_private_repo`: None
+- `hub_always_push`: False
+- `gradient_checkpointing`: False
+- `gradient_checkpointing_kwargs`: None
+- `include_inputs_for_metrics`: False
+- `include_for_metrics`: []
+- `eval_do_concat_batches`: True
+- `fp16_backend`: auto
+- `push_to_hub_model_id`: None
+- `push_to_hub_organization`: None
+- `mp_parameters`: 
+- `auto_find_batch_size`: False
+- `full_determinism`: False
+- `torchdynamo`: None
+- `ray_scope`: last
+- `ddp_timeout`: 1800
+- `torch_compile`: False
+- `torch_compile_backend`: None
+- `torch_compile_mode`: None
+- `dispatch_batches`: None
+- `split_batches`: None
+- `include_tokens_per_second`: False
+- `include_num_input_tokens_seen`: False
+- `neftune_noise_alpha`: None
+- `optim_target_modules`: None
+- `batch_eval_metrics`: False
+- `eval_on_start`: False
+- `use_liger_kernel`: False
+- `eval_use_gather_object`: False
+- `average_tokens_across_devices`: False
+- `prompts`: None
+- `batch_sampler`: batch_sampler
+- `multi_dataset_batch_sampler`: proportional
+
+</details>
+
+### Training Logs
+| Epoch      | Step     | Training Loss | gooaq-dev_ndcg@10    |
+|:----------:|:--------:|:-------------:|:--------------------:|
+| -1         | -1       | -             | 0.1489 (-0.4423)     |
+| 0.0001     | 1        | 1.328         | -                    |
+| 0.0221     | 200      | 1.1586        | -                    |
+| 0.0443     | 400      | 0.7765        | -                    |
+| 0.0664     | 600      | 0.651         | -                    |
+| 0.0885     | 800      | 0.6165        | -                    |
+| 0.1106     | 1000     | 0.6434        | 0.7674 (+0.1762)     |
+| 0.1328     | 1200     | 0.5952        | -                    |
+| 0.1549     | 1400     | 0.573         | -                    |
+| 0.1770     | 1600     | 0.5538        | -                    |
+| 0.1992     | 1800     | 0.5492        | -                    |
+| 0.2213     | 2000     | 0.5452        | 0.8095 (+0.2182)     |
+| 0.2434     | 2200     | 0.5325        | -                    |
+| 0.2655     | 2400     | 0.5178        | -                    |
+| 0.2877     | 2600     | 0.5233        | -                    |
+| 0.3098     | 2800     | 0.5079        | -                    |
+| 0.3319     | 3000     | 0.5084        | 0.8178 (+0.2266)     |
+| 0.3541     | 3200     | 0.5104        | -                    |
+| 0.3762     | 3400     | 0.5053        | -                    |
+| 0.3983     | 3600     | 0.4892        | -                    |
+| 0.4204     | 3800     | 0.4879        | -                    |
+| 0.4426     | 4000     | 0.4969        | 0.8260 (+0.2348)     |
+| 0.4647     | 4200     | 0.492         | -                    |
+| 0.4868     | 4400     | 0.4798        | -                    |
+| 0.5090     | 4600     | 0.4708        | -                    |
+| 0.5311     | 4800     | 0.4638        | -                    |
+| 0.5532     | 5000     | 0.4746        | 0.8286 (+0.2374)     |
+| 0.5753     | 5200     | 0.4467        | -                    |
+| 0.5975     | 5400     | 0.4615        | -                    |
+| 0.6196     | 5600     | 0.452         | -                    |
+| 0.6417     | 5800     | 0.4632        | -                    |
+| 0.6639     | 6000     | 0.4517        | 0.8290 (+0.2378)     |
+| 0.6860     | 6200     | 0.447         | -                    |
+| 0.7081     | 6400     | 0.4581        | -                    |
+| 0.7303     | 6600     | 0.4521        | -                    |
+| 0.7524     | 6800     | 0.4461        | -                    |
+| 0.7745     | 7000     | 0.4418        | 0.8372 (+0.2459)     |
+| 0.7966     | 7200     | 0.4279        | -                    |
+| 0.8188     | 7400     | 0.4136        | -                    |
+| 0.8409     | 7600     | 0.4163        | -                    |
+| 0.8630     | 7800     | 0.4099        | -                    |
+| 0.8852     | 8000     | 0.4156        | 0.8431 (+0.2518)     |
+| 0.9073     | 8200     | 0.4146        | -                    |
+| 0.9294     | 8400     | 0.4264        | -                    |
+| 0.9515     | 8600     | 0.4261        | -                    |
+| 0.9737     | 8800     | 0.4145        | -                    |
+| **0.9958** | **9000** | **0.4219**    | **0.8475 (+0.2562)** |
+| -1         | -1       | -             | 0.8475 (+0.2562)     |
+
+* The bold row denotes the saved checkpoint.
+
+### Framework Versions
+- Python: 3.11.10
+- Sentence Transformers: 3.5.0.dev0
+- Transformers: 4.49.0
+- PyTorch: 2.5.1+cu124
+- Accelerate: 1.2.0
+- Datasets: 2.21.0
+- Tokenizers: 0.21.0
+
+## Citation
+
+### BibTeX
+
+#### Sentence Transformers
+```bibtex
+@inproceedings{reimers-2019-sentence-bert,
+    title = "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks",
+    author = "Reimers, Nils and Gurevych, Iryna",
+    booktitle = "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing",
+    month = "11",
+    year = "2019",
+    publisher = "Association for Computational Linguistics",
+    url = "https://arxiv.org/abs/1908.10084",
+}
+```
+
+<!--
+## Glossary
+
+*Clearly define terms in order to be accessible across audiences.*
+-->
+
+<!--
+## Model Card Authors
+
+*Lists the people who create the model card, providing recognition and accountability for the detailed work that goes into its construction.*
+-->
+
+<!--
+## Model Card Contact
+
+*Provides a way for people who have updates to the Model Card, suggestions, or questions, to contact the Model Card authors.*
+-->

config.json

@@ -0,0 +1,60 @@
+{
+  "_name_or_path": "../models/NeoBERT",
+  "architectures": [
+    "NeoBERTForSequenceClassification"
+  ],
+  "auto_map": {
+    "AutoConfig": "model.NeoBERTConfig",
+    "AutoModel": "model.NeoBERTLMHead",
+    "AutoModelForMaskedLM": "model.NeoBERTLMHead",
+    "AutoModelForSequenceClassification": "chandar-lab/NeoBERT--model.NeoBERTForSequenceClassification"
+  },
+  "classifier_init_range": 0.02,
+  "decoder_init_range": 0.02,
+  "dim_head": 64,
+  "embedding_init_range": 0.02,
+  "hidden_size": 768,
+  "id2label": {
+    "0": "LABEL_0"
+  },
+  "intermediate_size": 3072,
+  "kwargs": {
+    "_commit_hash": null,
+    "architectures": [
+      "NeoBERTLMHead"
+    ],
+    "attn_implementation": null,
+    "auto_map": {
+      "AutoConfig": "model.NeoBERTConfig",
+      "AutoModel": "model.NeoBERTLMHead",
+      "AutoModelForMaskedLM": "model.NeoBERTLMHead",
+      "AutoModelForSequenceClassification": "chandar-lab/NeoBERT--model.NeoBERTForSequenceClassification"
+    },
+    "classifier_init_range": 0.02,
+    "dim_head": 64,
+    "kwargs": {
+      "classifier_init_range": 0.02,
+      "pretrained_model_name_or_path": "google-bert/bert-base-uncased",
+      "trust_remote_code": true
+    },
+    "model_type": "neobert",
+    "pretrained_model_name_or_path": "google-bert/bert-base-uncased",
+    "torch_dtype": "float32",
+    "transformers_version": "4.48.2",
+    "trust_remote_code": true
+  },
+  "label2id": {
+    "LABEL_0": 0
+  },
+  "max_length": 4096,
+  "model_type": "neobert",
+  "norm_eps": 1e-05,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 28,
+  "pad_token_id": 0,
+  "pretrained_model_name_or_path": "google-bert/bert-base-uncased",
+  "torch_dtype": "float32",
+  "transformers_version": "4.49.0",
+  "trust_remote_code": true,
+  "vocab_size": 30522
+}

model.py

@@ -0,0 +1,420 @@
+# From https://github.com/facebookresearch/llama/blob/main/llama/model.py
+
+import torch
+from torch import nn
+
+from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss
+from torch.nn.functional import scaled_dot_product_attention
+
+from typing import Optional
+import numpy as np
+
+from xformers.ops import SwiGLU
+
+try:
+    from flash_attn.flash_attn_interface import flash_attn_varlen_func
+
+    FLASH_ATTN_AVAILABLE = True
+except ImportError:
+    FLASH_ATTN_AVAILABLE = False
+
+from transformers import (
+    PreTrainedModel,
+    PretrainedConfig,
+    DataCollatorForLanguageModeling,
+)
+from transformers.modeling_outputs import (
+    BaseModelOutput,
+    MaskedLMOutput,
+    SequenceClassifierOutput,
+)
+
+from .rotary import precompute_freqs_cis, apply_rotary_emb
+
+
+class DataCollatorWithPacking(DataCollatorForLanguageModeling):
+    def __init__(self, pack_sequences=False, **kwargs):
+        super().__init__(**kwargs)
+        self.pack_sequences = pack_sequences
+
+    def __call__(self, batch):
+        if self.pack_sequences:
+            # Add position_ids if not present
+            if "position_ids" not in batch[0]:
+                for item in batch:
+                    item["position_ids"] = list(range(len(item["input_ids"])))
+
+            # Pack the sequences into a single list
+            input_ids_list = [item["input_ids"] for item in batch]
+            position_ids_list = [item["position_ids"] for item in batch]
+            seqlens = np.array([0] + [len(ids) for ids in input_ids_list])
+
+            packed_batch = {
+                "position_ids": np.concatenate(position_ids_list, axis=0),
+                "input_ids": np.concatenate(input_ids_list, axis=0),
+                "cu_seqlens": np.cumsum(seqlens),
+                "max_seqlen": max(seqlens),
+            }
+
+            batch = super().__call__([packed_batch])
+            batch["cu_seqlens"] = batch["cu_seqlens"].to(torch.int32).squeeze()
+        else:
+            batch = super().__call__(batch)
+            batch["attention_mask"] = batch["attention_mask"].to(torch.bool)
+
+        return batch
+
+
+class NeoBERTConfig(PretrainedConfig):
+    model_type = "neobert"
+
+    # All config parameters must have a default value.
+    def __init__(
+        self,
+        hidden_size: int = 768,
+        num_hidden_layers: int = 28,
+        num_attention_heads: int = 12,
+        intermediate_size: int = 3072,
+        embedding_init_range: float = 0.02,
+        decoder_init_range: float = 0.02,
+        norm_eps: float = 1e-06,
+        vocab_size: int = 30522,
+        pad_token_id: int = 0,
+        max_length: int = 1024,
+        **kwargs,
+    ):
+        super().__init__(**kwargs)
+
+        self.hidden_size = hidden_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        if hidden_size % num_attention_heads != 0:
+            raise ValueError("Hidden size must be divisible by the number of heads.")
+        self.dim_head = hidden_size // num_attention_heads
+        self.intermediate_size = intermediate_size
+        self.embedding_init_range = embedding_init_range
+        self.decoder_init_range = decoder_init_range
+        self.norm_eps = norm_eps
+        self.vocab_size = vocab_size
+        self.pad_token_id = pad_token_id
+        self.max_length = max_length
+        self.kwargs = kwargs
+
+
+class EncoderBlock(nn.Module):
+    """Transformer encoder block."""
+
+    def __init__(self, config: NeoBERTConfig):
+        super().__init__()
+
+        self.config = config
+
+        # Attention
+        self.qkv = nn.Linear(in_features=config.hidden_size, out_features=config.hidden_size * 3, bias=False)
+        self.wo = nn.Linear(in_features=config.hidden_size, out_features=config.hidden_size, bias=False)
+
+        # Feedforward network
+        multiple_of = 8
+        intermediate_size = int(2 * config.intermediate_size / 3)
+        intermediate_size = multiple_of * ((intermediate_size + multiple_of - 1) // multiple_of)
+        self.ffn = SwiGLU(config.hidden_size, intermediate_size, config.hidden_size, bias=False)
+
+        # Layer norms
+        self.attention_norm = nn.RMSNorm(config.hidden_size, config.norm_eps)
+        self.ffn_norm = nn.RMSNorm(config.hidden_size, config.norm_eps)
+
+    def forward(
+        self,
+        x: torch.Tensor,
+        attention_mask: torch.Tensor,
+        freqs_cis: torch.Tensor,
+        output_attentions: bool,
+        max_seqlen: int = None,
+        cu_seqlens: torch.Tensor = None,
+    ):
+        # Attention
+        attn_output, attn_weights = self._att_block(
+            self.attention_norm(x), attention_mask, freqs_cis, output_attentions, max_seqlen, cu_seqlens
+        )
+
+        # Residual
+        x = x + attn_output
+
+        # Feed-forward
+        x = x + self.ffn(self.ffn_norm(x))
+
+        return x, attn_weights
+
+    def _att_block(
+        self,
+        x: torch.Tensor,
+        attention_mask: torch.Tensor,
+        freqs_cis: torch.Tensor,
+        output_attentions: bool,
+        max_seqlen: int = None,
+        cu_seqlens: torch.Tensor = None,
+    ):
+        batch_size, seq_len, _ = x.shape
+
+        xq, xk, xv = self.qkv(x).view(batch_size, seq_len, self.config.num_attention_heads, self.config.dim_head * 3).chunk(3, axis=-1)
+
+        xq, xk = apply_rotary_emb(xq, xk, freqs_cis)
+
+        # Attn block
+        attn_weights = None
+
+        # Flash attention if the tensors are packed
+        if cu_seqlens is not None:
+            attn = flash_attn_varlen_func(
+                q=xq.squeeze(0),
+                k=xk.squeeze(0),
+                v=xv.squeeze(0),
+                cu_seqlens_q=cu_seqlens,
+                cu_seqlens_k=cu_seqlens,
+                max_seqlen_q=max_seqlen,
+                max_seqlen_k=max_seqlen,
+                dropout_p=0.0,
+                causal=False,
+            )
+        # Eager attention if attention weights are needed in the output
+        elif output_attentions:
+            attn_weights = xq.permute(0, 2, 1, 3) @ xk.permute(0, 2, 3, 1) / (xq.size(-1) ** 0.5)
+            if attention_mask is not None:
+                attn_weights = attn_weights * attention_mask
+            attn_weights = attn_weights.softmax(-1)
+            attn = attn_weights @ xv.permute(0, 2, 1, 3)
+            attn = attn.transpose(1, 2)
+        # Fall back to SDPA otherwise
+        else:
+            attn = scaled_dot_product_attention(
+                query=xq.transpose(1, 2),
+                key=xk.transpose(1, 2),
+                value=xv.transpose(1, 2),
+                attn_mask=attention_mask.bool(),
+                dropout_p=0,
+            ).transpose(1, 2)
+
+        return self.wo(attn.reshape(batch_size, seq_len, self.config.num_attention_heads * self.config.dim_head)), attn_weights
+
+
+class NeoBERTPreTrainedModel(PreTrainedModel):
+    config_class = NeoBERTConfig
+    _supports_cache_class = True
+
+    def _init_weights(self, module):
+        if isinstance(module, nn.Linear):
+            module.weight.data.uniform_(-self.config.decoder_init_range, self.config.decoder_init_range)
+        elif isinstance(module, nn.Embedding):
+            module.weight.data.uniform_(-self.config.embedding_init_range, self.config.embedding_init_range)
+
+
+class NeoBERT(NeoBERTPreTrainedModel):
+    config_class = NeoBERTConfig
+
+    def __init__(self, config: NeoBERTConfig):
+        super().__init__(config)
+
+        self.config = config
+
+        self.encoder = nn.Embedding(config.vocab_size, config.hidden_size, padding_idx=config.pad_token_id)
+
+        # Ensures freqs_cis is moved to the same devices as the model. Non-persistent buffers are not saved in the state_dict.
+        freqs_cis = precompute_freqs_cis(config.hidden_size // config.num_attention_heads, config.max_length)
+        self.register_buffer("freqs_cis", freqs_cis, persistent=False)
+
+        self.transformer_encoder = nn.ModuleList()
+        for _ in range(config.num_hidden_layers):
+            self.transformer_encoder.append(EncoderBlock(config))
+
+        self.layer_norm = nn.RMSNorm(config.hidden_size, config.norm_eps)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def forward(
+        self,
+        input_ids: torch.Tensor,
+        position_ids: torch.Tensor = None,
+        max_seqlen: int = None,
+        cu_seqlens: torch.Tensor = None,
+        attention_mask: torch.Tensor = None,
+        output_hidden_states: bool = False,
+        output_attentions: bool = False,
+        **kwargs,
+    ):
+        # Initialize
+        hidden_states, attentions = [], []
+
+        # Expand and repeat: (Batch, Length) -> (Batch, Heads, Length, Length)
+        if attention_mask is not None:
+            attention_mask = attention_mask.unsqueeze(1).unsqueeze(1).repeat(1, self.config.num_attention_heads, attention_mask.size(-1), 1)
+
+        # Checks to be done if inputs are packed sequences
+        if cu_seqlens is not None:
+            assert (
+                FLASH_ATTN_AVAILABLE
+            ), "Flash-attention is not available. Please ''pip install flash_attn'', or provide un-packed sequences."
+            assert not output_attentions, "Output attentions is not supported when sequences are packed."
+            assert max_seqlen is not None, "Missing max_seqlen. It must be provided when cu_seqlens are not None."
+            assert input_ids.shape[0] == 1, "Cumulative sequence lengths are provided but input_ids are not packed."
+            assert input_ids.is_cuda, "Packing uses an implementation of flash-attention and is only supported on GPU."
+
+        # RoPE
+        freqs_cis = self.freqs_cis[position_ids] if position_ids is not None else self.freqs_cis[: input_ids.shape[1]].unsqueeze(0)
+
+        # Embedding
+        x = self.encoder(input_ids)
+
+        # Transformer encoder
+        for layer in self.transformer_encoder:
+            x, attn = layer(x, attention_mask, freqs_cis, output_attentions, max_seqlen, cu_seqlens)
+            if output_hidden_states:
+                hidden_states.append(x)
+            if output_attentions:
+                attentions.append(attn)
+
+        # Final normalization layer
+        x = self.layer_norm(x)
+
+        # Return the output of the last hidden layer
+        return BaseModelOutput(
+            last_hidden_state=x,
+            hidden_states=hidden_states if output_hidden_states else None,
+            attentions=attentions if output_attentions else None,
+        )
+
+
+class NeoBERTLMHead(NeoBERTPreTrainedModel):
+    config_class = NeoBERTConfig
+
+    def __init__(self, config: NeoBERTConfig):
+        super().__init__(config)
+
+        self.config = config
+
+        self.model = NeoBERT(config)
+        self.decoder = nn.Linear(config.hidden_size, config.vocab_size)
+
+        self.post_init()
+
+    def forward(
+        self,
+        input_ids: torch.Tensor,
+        position_ids: torch.Tensor = None,
+        max_seqlen: int = None,
+        cu_seqlens: torch.Tensor = None,
+        attention_mask: torch.Tensor = None,
+        output_hidden_states: bool = False,
+        output_attentions: bool = False,
+        **kwargs,
+    ):
+
+        output = self.model.forward(
+            input_ids,
+            position_ids,
+            max_seqlen,
+            cu_seqlens,
+            attention_mask,
+            output_hidden_states,
+            output_attentions,
+        )
+        logits = self.decoder(output.last_hidden_state)
+
+        return MaskedLMOutput(
+            hidden_states=output.hidden_states if output_hidden_states else None,
+            attentions=output.attentions if output_attentions else None,
+            logits=logits,
+        )
+
+
+class NeoBERTForSequenceClassification(NeoBERTPreTrainedModel):
+    config_class = NeoBERTConfig
+
+    def __init__(self, config: NeoBERTConfig):
+        super().__init__(config)
+
+        self.config = config
+
+        self.num_labels = getattr(config, "num_labels", 2)
+        self.classifier_dropout = getattr(config, "classifier_dropout", 0.1)
+        self.classifier_init_range = getattr(config, "classifier_init_range", 0.02)
+
+        self.model = NeoBERT(config)
+
+        self.dense = nn.Linear(self.config.hidden_size, self.config.hidden_size)
+        self.dropout = nn.Dropout(self.classifier_dropout)
+        self.classifier = nn.Linear(self.config.hidden_size, self.num_labels)
+
+        self.post_init()
+
+    def _init_weights(self, module):
+        if isinstance(module, nn.Linear):
+            module.weight.data.normal_(mean=0.0, std=self.classifier_init_range)
+            if module.bias is not None:
+                module.bias.data.zero_()
+
+    def forward(
+        self,
+        input_ids: torch.Tensor,
+        position_ids: torch.Tensor = None,
+        max_seqlen: int = None,
+        cu_seqlens: torch.Tensor = None,
+        attention_mask: torch.Tensor = None,
+        output_hidden_states: bool = False,
+        output_attentions: bool = False,
+        labels: Optional[torch.Tensor] = None,
+        return_dict: Optional[bool] = None,
+    ):
+
+        output = self.model.forward(
+            input_ids,
+            position_ids,
+            max_seqlen,
+            cu_seqlens,
+            attention_mask,
+            output_hidden_states,
+            output_attentions,
+        )
+        hidden_states = output.last_hidden_state
+
+        x = hidden_states[:, 0, :]
+        x = self.dropout(x)
+        x = self.dense(x)
+        x = torch.tanh(x)
+        x = self.dropout(x)
+
+        logits = self.classifier(x)
+
+        loss = None
+        if labels is not None:
+            if self.config.problem_type is None:
+                if self.num_labels == 1:
+                    self.config.problem_type = "regression"
+                elif self.num_labels > 1 and (labels.dtype == torch.long or labels.dtype == torch.int):
+                    self.config.problem_type = "single_label_classification"
+                else:
+                    self.config.problem_type = "multi_label_classification"
+
+            if self.config.problem_type == "regression":
+                loss_fct = MSELoss()
+                if self.num_labels == 1:
+                    loss = loss_fct(logits.squeeze(), labels.squeeze())
+                else:
+                    loss = loss_fct(logits, labels)
+            elif self.config.problem_type == "single_label_classification":
+                loss_fct = CrossEntropyLoss()
+                loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
+            elif self.config.problem_type == "multi_label_classification":
+                loss_fct = BCEWithLogitsLoss()
+                loss = loss_fct(logits, labels)
+
+        if not return_dict:
+            result = (logits,)
+            return ((loss,) + result) if loss is not None else result
+
+        return SequenceClassifierOutput(
+            loss=loss,
+            logits=logits,
+            hidden_states=output.hidden_states if output_hidden_states else None,
+            attentions=output.attentions if output_attentions else None,
+        )

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1e79b9fccea0b7f46362b245105068cc9b4dcef5880ef562d483ddf382032298
+size 889047508

rotary.py

@@ -0,0 +1,61 @@
+# From https://github.com/facebookresearch/llama/blob/main/llama/model.py
+
+import torch
+from typing import Tuple
+
+
+def precompute_freqs_cis(dim: int, end: int, theta: float = 10000.0):
+    """
+    Precompute the frequency tensor for complex exponentials (cis) with given dimensions.
+
+    This function calculates a frequency tensor with complex exponentials using the given dimension 'dim'
+    and the end index 'end'. The 'theta' parameter scales the frequencies.
+    The returned tensor contains complex values in complex64 data type.
+
+    Args:
+        dim (int): Dimension of the frequency tensor.
+        end (int): End index for precomputing frequencies.
+        theta (float, optional): Scaling factor for frequency computation. Defaults to 10000.0.
+
+    Returns:
+        torch.Tensor: Precomputed frequency tensor with complex exponentials.
+    """
+
+    freqs = 1.0 / (theta ** (torch.arange(0, dim, 2)[: (dim // 2)].float() / dim))
+    t = torch.arange(end, device=freqs.device)
+    freqs = torch.outer(t, freqs).float()
+    return torch.polar(torch.ones_like(freqs), freqs)
+
+
+def reshape_for_broadcast(freqs_cis: torch.Tensor, x: torch.Tensor):
+    assert freqs_cis.shape[1:] == (x.shape[1], x.shape[-1])
+    return freqs_cis.contiguous().unsqueeze(2)
+
+
+def apply_rotary_emb(
+    xq: torch.Tensor,
+    xk: torch.Tensor,
+    freqs_cis: torch.Tensor,
+) -> Tuple[torch.Tensor, torch.Tensor]:
+    """
+    Apply rotary embeddings to input tensors using the given frequency tensor.
+
+    This function applies rotary embeddings to the given query 'xq' and key 'xk' tensors using the provided
+    frequency tensor 'freqs_cis'. The input tensors are reshaped as complex numbers, and the frequency tensor
+    is reshaped for broadcasting compatibility. The resulting tensors contain rotary embeddings and are
+    returned as real tensors.
+
+    Args:
+        xq (torch.Tensor): Query tensor to apply rotary embeddings.
+        xk (torch.Tensor): Key tensor to apply rotary embeddings.
+        freqs_cis (torch.Tensor): Precomputed frequency tensor for complex exponentials.
+
+    Returns:
+        Tuple[torch.Tensor, torch.Tensor]: Tuple of modified query tensor and key tensor with rotary embeddings.
+    """
+    xq_ = torch.view_as_complex(xq.float().reshape(*xq.shape[:-1], -1, 2))
+    xk_ = torch.view_as_complex(xk.float().reshape(*xk.shape[:-1], -1, 2))
+    freqs_cis = reshape_for_broadcast(freqs_cis, xq_)
+    xq_out = torch.view_as_real(xq_ * freqs_cis).flatten(3)
+    xk_out = torch.view_as_real(xk_ * freqs_cis).flatten(3)
+    return xq_out.type_as(xq), xk_out.type_as(xk)

special_tokens_map.json

@@ -0,0 +1,37 @@
+{
+  "cls_token": {
+    "content": "[CLS]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "mask_token": {
+    "content": "[MASK]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "[PAD]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "sep_token": {
+    "content": "[SEP]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "[UNK]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer_config.json

@@ -0,0 +1,61 @@
+{
+  "added_tokens_decoder": {
+    "0": {
+      "content": "[PAD]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "100": {
+      "content": "[UNK]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "101": {
+      "content": "[CLS]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "102": {
+      "content": "[SEP]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "103": {
+      "content": "[MASK]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "clean_up_tokenization_spaces": false,
+  "cls_token": "[CLS]",
+  "do_lower_case": true,
+  "extra_special_tokens": {},
+  "mask_token": "[MASK]",
+  "model_input_names": [
+    "input_ids",
+    "attention_mask"
+  ],
+  "model_max_length": 4096,
+  "pad_token": "[PAD]",
+  "sep_token": "[SEP]",
+  "strip_accents": null,
+  "tokenize_chinese_chars": true,
+  "tokenizer_class": "BertTokenizer",
+  "unk_token": "[UNK]",
+  "vocab_size": 30522
+}