|
|
--- |
|
|
library_name: transformers |
|
|
license: apache-2.0 |
|
|
--- |
|
|
|
|
|
# II-Medical-8B |
|
|
|
|
|
<div style="display: flex; justify-content: center;"> |
|
|
<img src="https://cdn-uploads.huggingface.co/production/uploads/6389496ff7d3b0df092095ed/73Y-oDmehp0eJ2HWrfn3V.jpeg" width="800"> |
|
|
</div> |
|
|
|
|
|
## I. Model Overview |
|
|
|
|
|
II-Medical-8B is the newest advanced large language model developed by Intelligent Internet, specifically engineered to enhance AI-driven medical reasoning. Following the positive reception of our previous [II-Medical-7B-Preview](https://huggingface.co/Intelligent-Internet/II-Medical-7B-Preview), this new iteration significantly advances the capabilities of medical question answering, |
|
|
|
|
|
## II. Training Methodology |
|
|
|
|
|
We collected and generated a comprehensive set of reasoning datasets for the medical domain and performed SFT fine-tuning on the **Qwen/Qwen3-8B** model. Following this, we further optimized the SFT model by training DAPO on a hard-reasoning dataset to boost performance. |
|
|
|
|
|
For SFT stage we using the hyperparameters: |
|
|
|
|
|
- Max Length: 16378. |
|
|
- Batch Size: 128. |
|
|
- Learning-Rate: 5e-5. |
|
|
- Number Of Epoch: 8. |
|
|
|
|
|
For RL stage we setup training with: |
|
|
|
|
|
- Max prompt length: 2048 tokens. |
|
|
- Max response length: 12288 tokens. |
|
|
- Overlong buffer: Enabled, 4096 tokens, penalty factor 1.0. |
|
|
- Clip ratios: Low 0.2, High 0.28. |
|
|
- Batch sizes: Train prompt 512, Generation prompt 1536, Mini-batch 32. |
|
|
- Responses per prompt: 16. |
|
|
- Temperature: 1.0, Top-p: 1.0, Top-k: -1 (vLLM rollout). |
|
|
- Learning rate: 1e-6, Warmup steps: 10, Weight decay: 0.1. |
|
|
- Loss aggregation: Token-mean. |
|
|
- Gradient clipping: 1.0. |
|
|
- Entropy coefficient: 0. |
|
|
|
|
|
## III. Evaluation Results |
|
|
|
|
|
Our II-Medical-8B model achieved a 40% score on [HealthBench](https://openai.com/index/healthbench/), a comprehensive open-source benchmark evaluating the performance and safety of large language models in healthcare. This performance is comparable to OpenAI's o1 reasoning model and GPT-4.5, OpenAI's largest and most advanced model to date. We provide a comparison to models available in ChatGPT below. |
|
|
|
|
|
 |
|
|
Detailed result for HealthBench can be found [here](https://huggingface.co/datasets/Intelligent-Internet/OpenAI-HealthBench-II-Medical-8B-GPT-4.1). |
|
|
|
|
|
 |
|
|
|
|
|
We evaluate on ten medical QA benchmarks include MedMCQA, MedQA, PubMedQA, medical related questions from MMLU-Pro and GPQA, small QA sets from Lancet and the New England |
|
|
Journal of Medicine, 4 Options and 5 Options splits from the MedBullets platform and MedXpertQA. |
|
|
|
|
|
| Model | MedMC | MedQA | PubMed | MMLU-P | GPQA | Lancet | MedB-4 | MedB-5 | MedX | NEJM | Avg | |
|
|
|--------------------------|-------|-------|--------|--------|------|--------|--------|--------|------|-------|-------| |
|
|
| [HuatuoGPT-o1-72B](https://huggingface.co/FreedomIntelligence/HuatuoGPT-o1-72B) | 76.76 | 88.85 | 79.90 | 80.46 | 64.36| 70.87 | 77.27 | 73.05 |23.53 |76.29 | 71.13 | |
|
|
| [QWQ 32B](https://huggingface.co/Qwen/QwQ-32B) | 69.73 | 87.03 | 88.5 | 79.86 | 69.17| 71.3 | 72.07 | 69.01 |24.98 |75.12 | 70.68 | |
|
|
| [Qwen2.5-7B-IT](https://huggingface.co/Qwen/Qwen2.5-7B-Instruct) | 56.56 | 61.51 | 71.3 | 61.17 | 42.56| 61.17 | 46.75 | 40.58 |13.26 |59.04 | 51.39 | |
|
|
| [HuatuoGPT-o1-8B](http://FreedomIntelligence/HuatuoGPT-o1-8B) | 63.97 | 74.78 | **80.10** | 63.71 | 55.38| 64.32 | 58.44 | 51.95 |15.79 |64.84 | 59.32 | |
|
|
| [Med-reason](https://huggingface.co/UCSC-VLAA/MedReason-8B) | 61.67 | 71.87 | 77.4 | 64.1 | 50.51| 59.7 | 60.06 | 54.22 |22.87 |66.8 | 59.92 | |
|
|
| [M1](https://huggingface.co/UCSC-VLAA/m1-7B-23K) | 62.54 | 75.81 | 75.80 | 65.86 | 53.08| 62.62 | 63.64 | 59.74 |19.59 |64.34 | 60.3 | |
|
|
| [II-Medical-8B-SFT](https://huggingface.co/II-Vietnam/II-Medical-8B-SFT) | **71.92** | 86.57 | 77.4 | 77.26 | 65.64| 69.17 | 76.30 | 67.53 |23.79 |**73.80** | 68.80 | |
|
|
| [II-Medical-8B](https://huggingface.co/Intelligent-Internet/II-Medical-8B) | 71.57 | **87.82** | 78.2 | **80.46** | **67.18**| **70.38** | **78.25** | **72.07** |**25.26** |73.13 | **70.49** | |
|
|
|
|
|
## IV. Dataset Curation |
|
|
|
|
|
The training dataset comprises 555,000 samples from the following sources: |
|
|
|
|
|
### 1. Public Medical Reasoning Datasets (103,031 samples) |
|
|
- [General Medical Reasoning](https://huggingface.co/datasets/GeneralReasoning/GeneralThought-430K): 40,544 samples |
|
|
- [Medical-R1-Distill-Data](https://huggingface.co/datasets/FreedomIntelligence/Medical-R1-Distill-Data): 22,000 samples |
|
|
- [Medical-R1-Distill-Data-Chinese](https://huggingface.co/datasets/FreedomIntelligence/Medical-R1-Distill-Data-Chinese): 17,000 samples |
|
|
- [UCSC-VLAA/m23k-tokenized](https://huggingface.co/datasets/UCSC-VLAA/m23k-tokenized): 23,487 samples |
|
|
|
|
|
### 2. Synthetic Medical QA Data with QwQ (225,700 samples) |
|
|
Generated from established medical datasets: |
|
|
- [MedMcQA](https://huggingface.co/datasets/openlifescienceai/medmcqa) (from openlifescienceai/medmcqa): 183,000 samples |
|
|
- [MedQA](https://huggingface.co/datasets/bigbio/med_qa): 10,000 samples |
|
|
- [MedReason](https://huggingface.co/datasets/UCSC-VLAA/MedReason): 32,700 samples |
|
|
|
|
|
### 3. Curated Medical R1 Traces (338,055 samples) |
|
|
|
|
|
First we gather all the public R1 traces from: |
|
|
|
|
|
- [PrimeIntellect/SYNTHETIC-1](https://huggingface.co/collections/PrimeIntellect/synthetic-1-67a2c399cfdd6c9f7fae0c37) |
|
|
- [GeneralReasoning/GeneralThought-430K](https://huggingface.co/datasets/GeneralReasoning/GeneralThought-430K) |
|
|
- [a-m-team/AM-DeepSeek-R1-Distilled-1.4M](https://arxiv.org/abs/2503.19633v1) |
|
|
- [open-thoughts/OpenThoughts2-1M](https://huggingface.co/datasets/open-thoughts/OpenThoughts2-1M) |
|
|
- [nvidia/Llama-Nemotron-Post-Training-Dataset](https://huggingface.co/datasets/nvidia/Llama-Nemotron-Post-Training-Dataset): Science subset only |
|
|
- Other resources: [cognitivecomputations/dolphin-r1](https://huggingface.co/datasets/cognitivecomputations/dolphin-r1), [ServiceNow-AI/R1-Distill-SFT](https://huggingface.co/datasets/ServiceNow-AI/R1-Distill-SFT),... |
|
|
|
|
|
All R1 reasoning traces were processed through a domain-specific pipeline as follows: |
|
|
|
|
|
1. Embedding Generation: Prompts are embedded using sentence-transformers/all-MiniLM-L6-v2. |
|
|
|
|
|
2. Clustering: Perform K-means clustering with 50,000 clusters. |
|
|
|
|
|
3. Domain Classification: |
|
|
|
|
|
- For each cluster, select the 10 prompts nearest to the cluster center. |
|
|
- Classify the domain of each selected prompt using Qwen2.5-32b-Instruct. |
|
|
- Assign the cluster's domain based on majority voting among the classified prompts. |
|
|
|
|
|
4. Domain Filtering: Keep only clusters labeled as Medical or Biology for the final dataset. |
|
|
|
|
|
|
|
|
### 4. Supplementary Math Dataset |
|
|
- Added 15,000 samples of reasoning traces from [light-r1](https://arxiv.org/abs/2503.10460) |
|
|
- Purpose: Enhance general reasoning capabilities of the model |
|
|
|
|
|
### Preprocessing Data |
|
|
1. Filtering for Complete Generation |
|
|
- Retained only traces with complete generation outputs |
|
|
|
|
|
2. Length-based Filtering |
|
|
- Minimum threshold: Keep only the prompt with more than 3 words. |
|
|
- Wait Token Filter: Removed traces with has more than 47 occurrences of "Wait" (97th percentile threshold). |
|
|
|
|
|
|
|
|
### Data Decontamination |
|
|
|
|
|
We using two step decontamination: |
|
|
1. Following [open-r1](https://github.com/huggingface/open-r1) project: We decontaminate a dataset using 10-grams with the evaluation datasets. |
|
|
2. After that, we using the fuzzy decontamination from [`s1k`](https://arxiv.org/abs/2501.19393) method with threshold 90%. |
|
|
|
|
|
**Our pipeline is carefully decontaminated with the evaluation datasets.** |
|
|
|
|
|
## V. How To Use |
|
|
Our model can be utilized in the same manner as Qwen or Deepseek-R1-Distill models. |
|
|
|
|
|
For instance, you can easily start a service using [vLLM](https://github.com/vllm-project/vllm): |
|
|
|
|
|
```bash |
|
|
vllm serve Intelligent-Internet/II-Medical-8B |
|
|
``` |
|
|
|
|
|
You can also easily start a service using [SGLang](https://github.com/sgl-project/sglang): |
|
|
|
|
|
```bash |
|
|
python -m sglang.launch_server --model Intelligent-Internet/II-Medical-8B |
|
|
``` |
|
|
|
|
|
## VI. Usage Guidelines |
|
|
|
|
|
- Recommended Sampling Parameters: temperature = 0.6, top_p = 0.9 |
|
|
- When using, explicitly request step-by-step reasoning and format the final answer within \boxed{} (e.g., "Please reason step-by-step, and put your final answer within \boxed{}."). |
|
|
## VII. Limitations and Considerations |
|
|
|
|
|
- Dataset may contain inherent biases from source materials |
|
|
- Medical knowledge requires regular updates |
|
|
- Please note that **It’s not suitable for medical use.** |
|
|
|
|
|
|
|
|
## VIII. Citation |
|
|
|
|
|
```bib |
|
|
@misc{2025II-Medical-8B, |
|
|
title={II-Medical-8B: Medical Reasoning Model}, |
|
|
author={Intelligent Internet}, |
|
|
year={2025} |
|
|
} |
|
|
``` |
