commit 3c68119d92303e5ae3519a14837b63d842261f0a Author: ModelHub XC Date: Sat Aug 8 09:37:17 2026 +0800 初始化项目，由ModelHub XC社区提供模型 Model: OmkarShewale/clintrial-qwen2.5-7b-sft Source: Original Platform diff --git a/.gitattributes b/.gitattributes new file mode 100644 index 0000000..52373fe --- /dev/null +++ b/.gitattributes @@ -0,0 +1,36 @@ +*.7z filter=lfs diff=lfs merge=lfs -text +*.arrow filter=lfs diff=lfs merge=lfs -text +*.bin filter=lfs diff=lfs merge=lfs -text +*.bz2 filter=lfs diff=lfs merge=lfs -text +*.ckpt filter=lfs diff=lfs merge=lfs -text +*.ftz filter=lfs diff=lfs merge=lfs -text +*.gz filter=lfs diff=lfs merge=lfs -text +*.h5 filter=lfs diff=lfs merge=lfs -text +*.joblib filter=lfs diff=lfs merge=lfs -text +*.lfs.* filter=lfs diff=lfs merge=lfs -text +*.mlmodel filter=lfs diff=lfs merge=lfs -text +*.model filter=lfs diff=lfs merge=lfs -text +*.msgpack filter=lfs diff=lfs merge=lfs -text +*.npy filter=lfs diff=lfs merge=lfs -text +*.npz filter=lfs diff=lfs merge=lfs -text +*.onnx filter=lfs diff=lfs merge=lfs -text +*.ot filter=lfs diff=lfs merge=lfs -text +*.parquet filter=lfs diff=lfs merge=lfs -text +*.pb filter=lfs diff=lfs merge=lfs -text +*.pickle filter=lfs diff=lfs merge=lfs -text +*.pkl filter=lfs diff=lfs merge=lfs -text +*.pt filter=lfs diff=lfs merge=lfs -text +*.pth filter=lfs diff=lfs merge=lfs -text +*.rar filter=lfs diff=lfs merge=lfs -text +*.safetensors filter=lfs diff=lfs merge=lfs -text +saved_model/**/* filter=lfs diff=lfs merge=lfs -text +*.tar.* filter=lfs diff=lfs merge=lfs -text +*.tar filter=lfs diff=lfs merge=lfs -text +*.tflite filter=lfs diff=lfs merge=lfs -text +*.tgz filter=lfs diff=lfs merge=lfs -text +*.wasm filter=lfs diff=lfs merge=lfs -text +*.xz filter=lfs diff=lfs merge=lfs -text +*.zip filter=lfs diff=lfs merge=lfs -text +*.zst filter=lfs diff=lfs merge=lfs -text +*tfevents* filter=lfs diff=lfs merge=lfs -text +tokenizer.json filter=lfs diff=lfs merge=lfs -text diff --git a/README.md b/README.md new file mode 100644 index 0000000..49b2251 --- /dev/null +++ b/README.md @@ -0,0 +1,135 @@ +--- +license: apache-2.0 +base_model: Qwen/Qwen2.5-7B-Instruct +library_name: transformers +pipeline_tag: text-generation +language: + - en +tags: + - clinical-trials + - healthcare + - biomedical + - information-extraction + - qlora + - peft + - trl + - grpo +--- + +# ClinTrial-LM — Qwen2.5-7B fine-tuned for clinical-trial understanding + +A **QLoRA**-fine-tuned [Qwen2.5-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-7B-Instruct) +specialised for clinical-trial tasks, trained on ~26k instruction examples derived from the +[ClinicalTrials.gov](https://clinicaltrials.gov) registry. + +📦 **Code, full pipeline & write-up:** https://github.com/omkar-droid/clintrial-lm + +> ⚕️ **Research/education only. Not medical advice.** Do not use for clinical decision-making. + +## What it does + +| Task | Description | +|---|---| +| **Eligibility extraction** | Free-text inclusion/exclusion criteria → structured JSON | +| **Plain-language summary** | Trial description → patient-friendly summary | +| **Condition Q&A** | "What conditions does this trial study?" | +| **Phase classification** | Identify the trial phase | + +## Results + +Evaluated on a **held-out test set split by trial ID** (no trial appears in both train and test), +500 sampled examples. + +| Task | Metric | Base Qwen2.5-7B | **This model (SFT)** | +|---|---|---|---| +| Eligibility extraction | criterion F1 | 0.840 | **0.968** | +| Eligibility output | **JSON validity** | 0.986 | **1.000** | +| Phase classification | exact match | 0.000 | **0.794** | +| Condition Q&A | token F1 | 0.298 | **0.742** | +| Plain-language summary | ROUGE-L | 0.195 | **0.290** | + +Eligibility extraction reaches **precision 0.962 / recall 0.978** with **100% schema-valid JSON**. + +The base model scores 0.000 on phase classification not because it lacks the knowledge, but because +it won't answer in the required format — fine-tuning buys **format discipline and faithfulness**. + +## Usage + +```python +import torch +from transformers import AutoModelForCausalLM, AutoTokenizer + +MODEL_ID = "OmkarShewale/clintrial-qwen2.5-7b-sft" +tokenizer = AutoTokenizer.from_pretrained(MODEL_ID) +model = AutoModelForCausalLM.from_pretrained(MODEL_ID, dtype=torch.bfloat16, device_map="auto") + +SYSTEM = ( + "You are a clinical research assistant. You help patients and clinicians understand " + "clinical trials. Answer only from the information provided, be precise, and never " + "invent eligibility criteria, conditions, or outcomes." +) + +criteria = """Inclusion Criteria: +1. Adults aged 18 years or older with confirmed type 2 diabetes +2. HbA1c between 7.0% and 10.5% at screening + +Exclusion Criteria: +1. History of severe hypoglycaemia within the last 6 months +2. Pregnancy or breastfeeding +""" + +messages = [ + {"role": "system", "content": SYSTEM}, + {"role": "user", "content": + 'Extract the eligibility criteria from the trial text below into JSON with two lists, ' + '"inclusion" and "exclusion". Copy each criterion verbatim; do not add any.\n\n' + f"Trial eligibility text:\n{criteria}"}, +] + +enc = tokenizer.apply_chat_template(messages, add_generation_prompt=True, + return_tensors="pt", return_dict=True).to(model.device) +out = model.generate(**enc, max_new_tokens=1024, do_sample=False) +print(tokenizer.decode(out[0, enc["input_ids"].shape[1]:], skip_special_tokens=True)) +``` + +**Note:** eligibility answers can be long (up to ~1.7k tokens). Use `max_new_tokens >= 1024` or the +JSON will be truncated and fail to parse. + +## Training + +| | | +|---|---| +| Base model | Qwen2.5-7B-Instruct | +| Method | QLoRA — 4-bit NF4 + LoRA (r=16, α=32, dropout 0.05) | +| Target modules | q/k/v/o_proj, gate/up/down_proj | +| Trainable params | ~0.5% of total | +| Precision | bf16 compute, gradient checkpointing | +| Effective batch | 32 (8 × 4 grad-accum) | +| LR / schedule | 2e-4, cosine, 3% warmup | +| Hardware | 1× NVIDIA H100 NVL (95 GB), ~19 GB used | +| Framework | HuggingFace TRL + PEFT + Transformers | + +**Best-checkpoint selection matters here:** eval loss bottomed around step 500 and then *rose* while +train loss kept falling (overfitting). `load_best_model_at_end` on `eval_loss` selected the step-500 +checkpoint — this model. Training for 3 epochs was more than necessary. + +A **GRPO / RLVR** variant (RL against a programmatic reward: JSON validity + criterion F1) was also +trained; it matched this SFT model but did not beat it, because SFT had already saturated the reward. +Details in the [repo](https://github.com/omkar-droid/clintrial-lm). + +## Data + +Built from 8,000 real ClinicalTrials.gov studies. Targets come from the registry's **own structured +fields** (not a teacher LLM), so labels are auditable. Splits are partitioned **by trial**, not by +example, to prevent leakage. + +## Limitations + +- **Not medical advice.** Outputs may be wrong; a human expert must review anything clinical. +- Trained on English registry text only; performance on other formats/languages is unknown. +- Summary quality (ROUGE-L 0.29) is the weakest task. +- Only 8k of ~500k available trials were used — more data would likely improve generalisation. + +## License + +Apache-2.0, inherited from Qwen2.5. Source registry data is public-domain U.S. government work. diff --git a/chat_template.jinja b/chat_template.jinja new file mode 100644 index 0000000..bdf7919 --- /dev/null +++ b/chat_template.jinja @@ -0,0 +1,54 @@ +{%- if tools %} + {{- '<|im_start|>system\n' }} + {%- if messages[0]['role'] == 'system' %} + {{- messages[0]['content'] }} + {%- else %} + {{- 'You are Qwen, created by Alibaba Cloud. You are a helpful assistant.' }} + {%- endif %} + {{- "\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within XML tags:\n" }} + {%- for tool in tools %} + {{- "\n" }} + {{- tool | tojson }} + {%- endfor %} + {{- "\n\n\nFor each function call, return a json object with function name and arguments within XML tags:\n\n{\"name\": , \"arguments\": }\n<|im_end|>\n" }} +{%- else %} + {%- if messages[0]['role'] == 'system' %} + {{- '<|im_start|>system\n' + messages[0]['content'] + '<|im_end|>\n' }} + {%- else %} + {{- '<|im_start|>system\nYou are Qwen, created by Alibaba Cloud. You are a helpful assistant.<|im_end|>\n' }} + {%- endif %} +{%- endif %} +{%- for message in messages %} + {%- if (message.role == "user") or (message.role == "system" and not loop.first) or (message.role == "assistant" and not message.tool_calls) %} + {{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>' + '\n' }} + {%- elif message.role == "assistant" %} + {{- '<|im_start|>' + message.role }} + {%- if message.content %} + {{- '\n' + message.content }} + {%- endif %} + {%- for tool_call in message.tool_calls %} + {%- if tool_call.function is defined %} + {%- set tool_call = tool_call.function %} + {%- endif %} + {{- '\n\n{"name": "' }} + {{- tool_call.name }} + {{- '", "arguments": ' }} + {{- tool_call.arguments | tojson }} + {{- '}\n' }} + {%- endfor %} + {{- '<|im_end|>\n' }} + {%- elif message.role == "tool" %} + {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != "tool") %} + {{- '<|im_start|>user' }} + {%- endif %} + {{- '\n\n' }} + {{- message.content }} + {{- '\n' }} + {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %} + {{- '<|im_end|>\n' }} + {%- endif %} + {%- endif %} +{%- endfor %} +{%- if add_generation_prompt %} + {{- '<|im_start|>assistant\n' }} +{%- endif %} diff --git a/config.json b/config.json new file mode 100644 index 0000000..98a8ea4 --- /dev/null +++ b/config.json @@ -0,0 +1,61 @@ +{ + "architectures": [ + "Qwen2ForCausalLM" + ], + "attention_dropout": 0.0, + "bos_token_id": 151643, + "dtype": "bfloat16", + "eos_token_id": 151645, + "hidden_act": "silu", + "hidden_size": 3584, + "initializer_range": 0.02, + "intermediate_size": 18944, + "layer_types": [ + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention" + ], + "max_position_embeddings": 32768, + "max_window_layers": 28, + "model_type": "qwen2", + "num_attention_heads": 28, + "num_hidden_layers": 28, + "num_key_value_heads": 4, + "pad_token_id": null, + "rms_norm_eps": 1e-06, + "rope_parameters": { + "rope_theta": 1000000.0, + "rope_type": "default" + }, + "sliding_window": null, + "tie_word_embeddings": false, + "transformers_version": "5.13.1", + "use_cache": true, + "use_sliding_window": false, + "vocab_size": 152064 +} diff --git a/generation_config.json b/generation_config.json new file mode 100644 index 0000000..a866a53 --- /dev/null +++ b/generation_config.json @@ -0,0 +1,14 @@ +{ + "bos_token_id": 151643, + "do_sample": true, + "eos_token_id": [ + 151645, + 151643 + ], + "pad_token_id": 151643, + "repetition_penalty": 1.05, + "temperature": 0.7, + "top_k": 20, + "top_p": 0.8, + "transformers_version": "5.13.1" +} diff --git a/model.safetensors b/model.safetensors new file mode 100644 index 0000000..3d1e2be --- /dev/null +++ b/model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0e1e5c534f46bdb2d27434016f1c62f07a8aba7b1a66483d06b22e0f9ffbba4b +size 15231272152 diff --git a/tokenizer.json b/tokenizer.json new file mode 100644 index 0000000..34510ff --- /dev/null +++ b/tokenizer.json @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3fd169731d2cbde95e10bf356d66d5997fd885dd8dbb6fb4684da3f23b2585d8 +size 11421892 diff --git a/tokenizer_config.json b/tokenizer_config.json new file mode 100644 index 0000000..770e41d --- /dev/null +++ b/tokenizer_config.json @@ -0,0 +1,30 @@ +{ + "add_prefix_space": false, + "backend": "tokenizers", + "bos_token": null, + "clean_up_tokenization_spaces": false, + "eos_token": "<|im_end|>", + "errors": "replace", + "extra_special_tokens": [ + "<|im_start|>", + "<|im_end|>", + "<|object_ref_start|>", + "<|object_ref_end|>", + "<|box_start|>", + "<|box_end|>", + "<|quad_start|>", + "<|quad_end|>", + "<|vision_start|>", + "<|vision_end|>", + "<|vision_pad|>", + "<|image_pad|>", + "<|video_pad|>" + ], + "is_local": false, + "local_files_only": false, + "model_max_length": 131072, + "pad_token": "<|endoftext|>", + "split_special_tokens": false, + "tokenizer_class": "Qwen2Tokenizer", + "unk_token": null +}