mikalv/gpt-oss-20b-chemistry

GPT-OSS-20B Finetuned for Chemistry

This model is a finetuned version of openai/gpt-oss-20b specialized for chemistry and scientific knowledge.

Training Details

• Base Model: openai/gpt-oss-20b
• Training Data: 99 examples (chemistry/science domain)
• LoRA Rank: 8
• Epochs: 10
• Hardware: A100 80GB GPU

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

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

prompt = "Hvordan fester man -OH til en benzenring?"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0]))

Quantized Version (GGUF)

For running on Mac M1/M2, use the GGUF quantized version:

• Q4_K_M quantization - Recommended for 16GB+ RAM
• Q5_K_M quantization - Best quality for 32GB+ RAM

Training Hyperparameters

• Learning Rate: 2e-4
• Batch Size: 1 (effective: 16 with gradient accumulation)
• LoRA Alpha: 16
• LoRA Dropout: 0.05
• Warmup Steps: 10

License

Same as base model (Apache 2.0)

Citation

@misc{gpt-oss-20b-chemistry,
  author = {Your Name},
  title = {GPT-OSS-20B Finetuned for Chemistry},
  year = {2024},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/YOUR_USERNAME/gpt-oss-20b-chemistry}}
}