Edit model card

This model is a PEFT finetune on google/gemma-1.1-2b-it. It's primary purpose is to be a single-turn assistance on topics of physiology

Training was done for 3 epochs on sardukar/physiology-mcqa-8k dataset.

Prompt

Prompt follows the chat template of Gemma-1.1-2b-it and should start with the phrase "You're an expert physician with deep knowledge of physiology\n{user_prompt}".

Inference parameters

input_text = [{"role": "user", "content": "You're an expert physician with deep knowledge of physiology\nWhat effect does GLP-1?"}]

with torch.inference_mode():
    input_ids = tokenizer.apply_chat_template(input_text, return_tensors="pt", add_generation_prompt=True).to("cuda")
    
    outputs = model.generate(
        input_ids,
        do_sample=True,
        use_cache=True,
        repetition_penalty=1.2,
        max_new_tokens=512,
        temperature=0.5,
        top_p=0.95,
        top_k=100,
        num_beams=4,
        early_stopping = True,
        return_dict_in_generate=True,
        output_attentions=False,
        output_hidden_states=False,
        output_scores=False,
        pad_token_id=tokenizer.pad_token_id,
        eos_token_id=tokenizer.eos_token_id
    )

Results

Below are the cherrypicked results of a training:

  • Prompt: You're an expert physician with deep knowledge of physiology\nExplain what is normal anion gap metabolic acidosis
  • Response:
Decreased HCO3- secretion by the kidney

Explanation: An anion gap metabolic acidosis occurs when the HCO3- concentration of the plasma is normal, but there is a net loss of anions from the body. This can occur if there is a decrease in the rate of HCO3- secretion by the kidney, or if there is an increase in the rate of HCO3- reabsorption by the tubules. Ref: Ganong's Review of Medical Physiology 23rd edition, Chapter 36.
  • Prompt: You're an expert physician with deep knowledge of physiology\nWhat effect does GLP-1?
  • Response:
Increases insulin secretion from beta cells

Explanation: GLP-1 increases insulin secretion from beta cells. Ref: Ganong's Review of Medical Physiology 23rd edition, Chapter 35.
  • Prompt: You're an expert physician with deep knowledge of physiology\nWhere Pacemaker cells of respiration are located?
  • Response:
Pre Botzinger Complex

Explanation: Pacemaker cells of respiration are located in Pre Botzinger Complex Ref: guyton and hall textbook of medical physiology 12 edition page number: 370,371,372
Downloads last month
9
Safetensors
Model size
2.51B params
Tensor type
FP16
·
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Dataset used to train sardukar/physiogemma-2b