Text Generation
Transformers
Safetensors
English
llama
mergekit
Merge
medical
conversational
Eval Results
Inference Endpoints
text-generation-inference
Medichat-Llama3-8B / README.md
sethuiyer's picture
Update README.md
534838c verified
metadata
base_model:
  - Undi95/Llama-3-Unholy-8B
  - Locutusque/llama-3-neural-chat-v1-8b
  - ruslanmv/Medical-Llama3-8B-16bit
library_name: transformers
tags:
  - mergekit
  - merge
  - medical
license: other
datasets:
  - mlabonne/orpo-dpo-mix-40k
  - Open-Orca/SlimOrca-Dedup
  - jondurbin/airoboros-3.2
  - microsoft/orca-math-word-problems-200k
  - m-a-p/Code-Feedback
  - MaziyarPanahi/WizardLM_evol_instruct_V2_196k
  - ruslanmv/ai-medical-chatbot
model-index:
  - name: Medichat-Llama3-8B
    results:
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: AI2 Reasoning Challenge (25-Shot)
          type: ai2_arc
          config: ARC-Challenge
          split: test
          args:
            num_few_shot: 25
        metrics:
          - type: acc_norm
            value: 59.13
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=sethuiyer/Medichat-Llama3-8B
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: HellaSwag (10-Shot)
          type: hellaswag
          split: validation
          args:
            num_few_shot: 10
        metrics:
          - type: acc_norm
            value: 82.9
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=sethuiyer/Medichat-Llama3-8B
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MMLU (5-Shot)
          type: cais/mmlu
          config: all
          split: test
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 60.35
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=sethuiyer/Medichat-Llama3-8B
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: TruthfulQA (0-shot)
          type: truthful_qa
          config: multiple_choice
          split: validation
          args:
            num_few_shot: 0
        metrics:
          - type: mc2
            value: 49.65
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=sethuiyer/Medichat-Llama3-8B
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: Winogrande (5-shot)
          type: winogrande
          config: winogrande_xl
          split: validation
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 78.93
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=sethuiyer/Medichat-Llama3-8B
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: GSM8k (5-shot)
          type: gsm8k
          config: main
          split: test
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 60.35
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=sethuiyer/Medichat-Llama3-8B
          name: Open LLM Leaderboard
language:
  - en

Medichat-Llama3-8B

Built upon the powerful LLaMa-3 architecture and fine-tuned on an extensive dataset of health information, this model leverages its vast medical knowledge to offer clear, comprehensive answers.

This model is generally better for accurate and informative responses, particularly for users seeking in-depth medical advice.

The following YAML configuration was used to produce this model:


models:
  - model: Undi95/Llama-3-Unholy-8B
    parameters:
      weight: [0.25, 0.35, 0.45, 0.35, 0.25]
      density: [0.1, 0.25, 0.5, 0.25, 0.1]
  - model: Locutusque/llama-3-neural-chat-v1-8b
  - model: ruslanmv/Medical-Llama3-8B-16bit
    parameters:
      weight: [0.55, 0.45, 0.35, 0.45, 0.55]
      density: [0.1, 0.25, 0.5, 0.25, 0.1]
merge_method: dare_ties
base_model: Locutusque/llama-3-neural-chat-v1-8b
parameters:
  int8_mask: true
dtype: bfloat16

Comparision Against Dr.Samantha 7B

Subject Medichat-Llama3-8B Accuracy (%) Dr. Samantha Accuracy (%)
Clinical Knowledge 71.70 52.83
Medical Genetics 78.00 49.00
Human Aging 70.40 58.29
Human Sexuality 73.28 55.73
College Medicine 62.43 38.73
Anatomy 64.44 41.48
College Biology 72.22 52.08
High School Biology 77.10 53.23
Professional Medicine 63.97 38.73
Nutrition 73.86 50.33
Professional Psychology 68.95 46.57
Virology 54.22 41.57
High School Psychology 83.67 66.60
Average 70.33 48.85

The current model demonstrates a substantial improvement over the previous Dr. Samantha model in terms of subject-specific knowledge and accuracy.

Usage:

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

class MedicalAssistant:
    def __init__(self, model_name="sethuiyer/Medichat-Llama3-8B", device="cuda"):
        self.device = device
        self.tokenizer = AutoTokenizer.from_pretrained(model_name)
        self.model = AutoModelForCausalLM.from_pretrained(model_name).to(self.device)
        self.sys_message = ''' 
        You are an AI Medical Assistant trained on a vast dataset of health information. Please be thorough and
        provide an informative answer. If you don't know the answer to a specific medical inquiry, advise seeking professional help.
        '''

    def format_prompt(self, question):
        messages = [
            {"role": "system", "content": self.sys_message},
            {"role": "user", "content": question}
        ]
        prompt = self.tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
        return prompt

    def generate_response(self, question, max_new_tokens=512):
        prompt = self.format_prompt(question)
        inputs = self.tokenizer(prompt, return_tensors="pt").to(self.device)
        with torch.no_grad():
            outputs = self.model.generate(**inputs, max_new_tokens=max_new_tokens, use_cache=True)
        answer = self.tokenizer.batch_decode(outputs, skip_special_tokens=True)[0].strip()
        return answer

if __name__ == "__main__":
    assistant = MedicalAssistant()
    question = '''
    Symptoms:
    Dizziness, headache, and nausea.

    What is the differential diagnosis?
    '''
    response = assistant.generate_response(question)
    print(response)

Quants

Thanks to Quant Factory, the quantized version of this model is available at QuantFactory/Medichat-Llama3-8B-GGUF,

Ollama

This model is now also available on Ollama. You can use it by running the command ollama run monotykamary/medichat-llama3 in your terminal. If you have limited computing resources, check out this video to learn how to run it on a Google Colab backend.