metadata
library_name: transformers
tags:
- unsloth
- trl
- sft
license: mit
datasets:
- syubraj/medical-chat-phi-3.5-instruct-1k
language:
- en
base_model:
- microsoft/Phi-3.5-mini-instruct
pipeline_tag: text-generation
Model Card for Model ID
Model Details
🏥 Phi-3.5 Mini Instruct Medical Chat (Unsloth)
The MedicalChat-Phi-3.5-mini-instruct- fine-tuned model is designed to simulate doctor-patient conversations, offering medical consultations and suggestions based on patient queries. However, its accuracy may be limited in real-world scenarios, as the training dataset was relatively small.
🔍 Model Overview
- 🧑⚕️ Developed by: syubraj
- 📜 Model Type: Causal Language Model (CausalLM)
- 🗣️ Language: English
- 📜 License: MIT License
- 🛠️ Fine-tuned From: microsoft/phi-3.5-mini-instruct
Uses
Direct Use
!pip install unsloth
from unsloth import FastLanguageModel
model, tokenizer = FastLanguageModel.from_pretrained("syubraj/Phi3.5-medicalchat-unsloth",
max_seq_length = 1024,
load_in_4bit = True,
dtype = None
)
user_query = "<Your medical query here>"
system_prompt = """You are a trusted AI-powered medical assistant. Analyze patient queries carefully and provide accurate, professional, and empathetic responses. Prioritize patient safety, adhere to medical best practices, and recommend consulting a healthcare provider when necessary."""
message = [
{"role": "system", "content": system_prompt},
{"role": "human", "content": user_query}
]
# Creating message based on tokenizers chat template
prompt = tokenizer.apply_chat_template(message, tokenize = False, add_generation_prompt = True)
FastLanguageModel.for_inference(model)
# Tokenizing inputs
inputs = tokenizer(prompt, return_tensors = "pt").to("cuda")
# Output Generated
outputs = model.generate(**inputs, max_new_tokens=256, use_cache=True) # Change the `max_new_tokens` according to required objective
tokenizer.batch_decode(outputs)
Model Results
| Step | Training Loss |
|---|---|
| 10 | 2.53 |
| 20 | 2.20 |
| 30 | 1.95 |
| 40 | 2.01 |
| 50 | 1.97 |
| 60 | 2.02 |