Tonic/med-gpt-oss-20b

med-gpt-oss-20b

A fine-tuned version of OpenAI's GPT-OSS-20B model for medical reasoning and instruction following.

Model Details

• —Base Model: openai/gpt-oss-20B
• —Model Type: Causal Language Model
• —Languages: English
• —License: Apache 2.0

Usage

Main Model

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

# Load the main model
model = AutoModelForCausalLM.from_pretrained(
    "Tonic/med-gpt-oss-20b",
    device_map="auto",
    torch_dtype=torch.bfloat16
)
tokenizer = AutoTokenizer.from_pretrained("Tonic/med-gpt-oss-20b")

# Generate text
input_text = "What are we having for dinner?"
input_ids = tokenizer(input_text, return_tensors="pt").to(model.device.type)
output = model.generate(**input_ids, max_new_tokens=50)
print(tokenizer.decode(output[0], skip_special_tokens=True))

Training Information

Training Configuration

• —Base Model: openai/gpt-oss-20b
• —Dataset: FreedomIntelligence/medical-o1-reasoning-SFT
• —Training Config: GPT-OSS Configuration
• —Trainer Type: SFTTrainer

Training Parameters

• —Batch Size: 4
• —Gradient Accumulation: 16
• —Learning Rate: 2e-4
• —Max Epochs: 1
• —Sequence Length: 2048

Training Infrastructure

• —Hardware: GPU (H100/A100)
• —Monitoring: Trackio integration
• —Experiment: med-track

Model Architecture

This is a fine-tuned version of the SmolLM3-3B model with the following specifications:

• —Base Model: SmolLM3-3B
• —Parameters: ~3B
• —Context Length: 2048
• —Languages: English, French
• —Architecture: Transformer-based causal language model

Performance

The model provides:

• —Medical Reasoning: High-quality medical reasoning
• —Conversation: Medical instruction following

Limitations

1. 1.Context Length: Limited by the model's maximum sequence length
2. 2.Bias: May inherit biases from the training data
3. 3.Factual Accuracy: May generate incorrect or outdated information
4. 4.Safety: Should be used responsibly with appropriate safeguards

Training Data

The model was fine-tuned on:

• —Dataset: FreedomIntelligence/medical-o1-reasoning-SFT
• —Size: ~20K samples
• —Format: reasoning
• —Languages: English

Monitoring

Citation

If you use this model in your research, please cite:

@misc{med_gpt_oss_20B,
  title={{med-gpt-oss-20b}},
  author={Joseph "Tonic" Pollack},
  year={2024},
  url={https://huggingface.co/Tonic/med-gpt-oss-20b}
}

License

This model is licensed under the Apache 2.0 License.

Acknowledgments

• —Base Model: SmolLM3-3B by HuggingFaceTB
• —Training Framework: PyTorch, Transformers, PEFT
• —Monitoring: Trackio integration
• —Quantization: torchao library

Support

For questions and support:

• —Open an issue on the Hugging Face repository
• —Check the model documentation
• —Review the training logs and configuration

Repository Structure

Tonic/med-gpt-oss-20b/
├── README.md (this file)
├── config.json
├── pytorch_model.bin
├── tokenizer.json
└── tokenizer_config.json

Usage Examples

Text Generation

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("Tonic/med-gpt-oss-20b")
tokenizer = AutoTokenizer.from_pretrained("Tonic/med-gpt-oss-20b")

text = "The future of artificial intelligence is"
inputs = tokenizer(text, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=100)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Conversation

def chat_with_model(prompt, max_length=100):
    inputs = tokenizer(prompt, return_tensors="pt")
    outputs = model.generate(**inputs, max_new_tokens=max_length)
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

response = chat_with_model("Hello, how are you today?")
print(response)

Advanced Usage

# With generation parameters
outputs = model.generate(
    **inputs,
    max_new_tokens=100,
    temperature=0.7,
    top_p=0.9,
    do_sample=True,
    pad_token_id=tokenizer.eos_token_id
)

Monitoring and Tracking

This model was trained with comprehensive monitoring:

• —Trackio Space: N/A
• —Experiment: med-track
• —Dataset Repository: https://huggingface.co/datasets/HuggingFaceH4/Multilingual-Thinking
• —Training Logs: Available in the experiment data

Deployment

Requirements

pip install torch transformers accelerate

Hardware Requirements

• —Main Model: GPU with 8GB+ VRAM recommended

Changelog

• —v1.0.0: Initial release with fine-tuned model