update reqs

app.py

@@ -1,64 +1,114 @@
+import torch
 import gradio as gr
-from huggingface_hub import InferenceClient
+from peft import PeftModel
+from unsloth import FastLanguageModel
 
-"""
-For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
-"""
-client = InferenceClient("jason-moore/deepseek-soap")
+# Load the model using Unsloth's method
+def load_model():
+    # Base model that was used for fine-tuning
+    base_model_id = "unsloth/DeepSeek-R1-Distill-Llama-8B"
+    
+    # Load the base model using Unsloth
+    max_seq_length = 2048
+    base_model, tokenizer = FastLanguageModel.from_pretrained(
+        model_name=base_model_id,
+        max_seq_length=max_seq_length,
+        dtype=torch.float16,
+        load_in_4bit=False,  # Avoid 4-bit to reduce complexity for testing
+    )
+    
+    # Load the LoRA adapter from HuggingFace
+    lora_model_id = "your-username/your-model-repo"  # Replace with your HF model path
+    model = PeftModel.from_pretrained(base_model, lora_model_id)
+    
+    # Optimize for inference
+    FastLanguageModel.for_inference(model)
+    
+    return model, tokenizer
 
+# Function to generate SOAP notes
+def generate_soap_note(doctor_patient_conversation):
+    if not doctor_patient_conversation.strip():
+        return "Please enter a doctor-patient conversation."
+    
+    # Format prompt identical to how it was done during training
+    prompt = """Below is an instruction that describes a task, paired with an input that provides further context.
+Write a response that appropriately completes the request. Pay special attention to the format of the response.
 
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
+### Instruction:
+You are a medical expert with advanced knowledge in clinical reasoning, diagnostics, and treatment planning.
+Summarize the following medical conversation between Doctor and Patient into a SOAP note with the following structure:
 
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
+SUBJECTIVE: This section focuses on the patient's perspective, including their chief complaint, symptoms, and any relevant personal or medical history.
 
-    messages.append({"role": "user", "content": message})
+OBJECTIVE: This section contains factual, measurable observations and data
+collected during the encounter, such as vital signs, test results, and physical exam findings.
+Only include information actually present in the conversation
 
-    response = ""
+ASSESSMENT: This section involves the healthcare provider's analysis and
+interpretation of the subjective and objective data, leading to a diagnosis or a proposed problem.
 
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
+PLAN: This section outlines the next steps in the patient's care, including treatment recommendations, follow-up plans, or referrals.
 
-        response += token
-        yield response
+### Conversation:
+{}
 
+### Response:
+{}"""
+    
+    # Use the same formatting pattern you used during inference
+    formatted_prompt = prompt.format(doctor_patient_conversation, "")
+    
+    # Tokenize using your pattern
+    inputs = tokenizer([formatted_prompt], return_tensors="pt").to(model.device)
+    
+    # Generate using the same parameters you used
+    outputs = model.generate(
+        input_ids=inputs.input_ids,
+        attention_mask=inputs.attention_mask,
+        max_new_tokens=1200,
+        use_cache=True,
+        temperature=0.1,
+        top_p=0.95,
+    )
+    
+    # Decode and extract the response part
+    response = tokenizer.batch_decode(outputs)[0]
+    soap_note = response.split("### Response:")[1].strip()
+    
+    return soap_note
 
+# Load model and tokenizer (this will run once when the app starts)
+model, tokenizer = load_model()
+
+# Sample conversation for the example
+sample_conversation = """
+Doctor: Good morning, how are you feeling today?
+Patient: Not so great, doctor. I've had this persistent cough for about two weeks now.
+Doctor: I'm sorry to hear that. Can you tell me more about the cough? Is it dry or are you coughing up anything?
+Patient: It started as a dry cough, but for the past few days I've been coughing up some yellowish phlegm.
+Doctor: Do you have any other symptoms like fever, chills, or shortness of breath?
+Patient: I had a fever of 100.5°F two days ago. I've been feeling more tired than usual, and sometimes it's a bit hard to catch my breath after coughing a lot.
 """
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
-demo = gr.ChatInterface(
-    respond,
-    additional_inputs=[
-        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
-        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
-        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
-    ],
-)
 
+# Create Gradio interface
+demo = gr.Interface(
+    fn=generate_soap_note,
+    inputs=gr.Textbox(
+        lines=15, 
+        placeholder="Enter doctor-patient conversation here...",
+        label="Doctor-Patient Conversation",
+        value=sample_conversation
+    ),
+    outputs=gr.Textbox(
+        label="Generated SOAP Note", 
+        lines=15
+    ),
+    title="Medical SOAP Note Generator",
+    description="Enter a doctor-patient conversation to generate a structured SOAP note using a fine-tuned DeepSeek-R1-Distill-Llama-8B model.",
+    examples=[[sample_conversation]],
+    allow_flagging="never"
+)
 
-if __name__ == "__main__":
-    demo.launch()
+# Launch the app
+demo.launch()

requirements.txt

@@ -1 +1,6 @@
-huggingface_hub==0.25.2
+torch>=2.0.0
+transformers>=4.36.0
+peft>=0.6.0
+gradio>=3.50.0
+accelerate>=0.25.0
+unsloth>=2023.11.28