Hugging Face's logo

Spaces:
CodCodingCode
/
medical-test
Paused

App Files Community
medical-test
File size: 4,697 Bytes 
6b29344
 
 
 
 
 
65aa5a5
 
 
 
 
 
 
6b29344
 
65aa5a5
6b29344
 
65aa5a5
 
6b29344
65aa5a5
6b29344
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
 
import os
import torch
from huggingface_hub import snapshot_download
from transformers import AutoTokenizer, AutoModelForCausalLM
import gradio as gr

# ——— CONFIG ———
REPO_ID = "CodCodingCode/llama-3.1-8b-clinical"
SUBFOLDER = "checkpoint-45000"
HF_TOKEN = os.getenv("HUGGINGFACE_HUB_TOKEN")

# ——— DOWNLOAD ONLY THE CHECKPOINT FILES ———
local_cache = snapshot_download(
    repo_id=REPO_ID,
    token=HF_TOKEN,
    allow_patterns=[f"{SUBFOLDER}/*.json", f"{SUBFOLDER}/*.safetensors"],
)

# ——— POINT AT THE REAL FILES ———
MODEL_DIR = os.path.join(local_cache, SUBFOLDER)

# ——— LOAD MODEL & TOKENIZER ———
tokenizer = AutoTokenizer.from_pretrained(
    MODEL_DIR,
    use_fast=False,
    trust_remote_code=True,
)
model = AutoModelForCausalLM.from_pretrained(
    MODEL_DIR,
    device_map="auto",
    torch_dtype=torch.float16,
    trust_remote_code=True,
)
model.eval()


# === Role Agent with instruction/input/output format ===
class RoleAgent:
    def __init__(self, role_instruction):
        self.role_instruction = role_instruction

    def act(self, input_text):
        prompt = (
            f"Instruction: {self.role_instruction}\n"
            f"Input: {input_text}\n"
            f"Output:"
        )
        inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
        outputs = model.generate(
            **inputs,
            max_new_tokens=256,
            do_sample=True,
            temperature=0.7,
            pad_token_id=tokenizer.eos_token_id,
        )
        response = tokenizer.decode(outputs[0], skip_special_tokens=True)

        # extract THINKING / ANSWER if present
        thinking, answer = "", response
        if "THINKING:" in response and "ANSWER:" in response and "END" in response:
            block = response.split("THINKING:")[1].split("END")[0]
            thinking = block.split("ANSWER:")[0].strip()
            answer = block.split("ANSWER:")[1].strip()

        return {
            "instruction": f"You are {self.role_instruction}.",
            "input": input_text,
            "thinking": thinking,
            "output": answer,
        }


# === Agents ===
summarizer = RoleAgent(
    "You are a clinical summarizer trained to extract structured vignettes from doctor–patient dialogues."
)
diagnoser = RoleAgent(
    "You are a board-certified diagnostician that diagnoses patients."
)
questioner = RoleAgent("You are a physician asking questions to diagnose a patient.")

treatment_agent = RoleAgent(
    "You are a board-certified clinician. Based on the diagnosis and patient vignette provided below, suggest a concise treatment plan that could realistically be initiated by a primary care physician or psychiatrist."
)


# === Inference State ===
conversation_history = []
summary = ""
diagnosis = ""


# === Gradio Inference ===
def simulate_interaction(user_input, iterations=1):
    history = [f"Doctor: What brings you in today?", f"Patient: {user_input}"]
    summary, diagnosis = "", ""

    for i in range(iterations):
        # Summarize
        sum_in = "\n".join(history) + f"\nPrevious Vignette: {summary}"
        sum_out = summarizer.act(sum_in)
        summary = sum_out["output"]

        # Diagnose
        diag_out = diagnoser.act(summary)
        diagnosis = diag_out["output"]

        # Question
        q_in = f"Vignette: {summary}\nCurrent Estimated Diagnosis: {diag_out['thinking']} {diagnosis}"
        q_out = questioner.act(q_in)
        history.append(f"Doctor: {q_out['output']}")
        history.append("Patient: (awaiting response)")

        # Treatment
        treatment_out = treatment_agent.act(
            f"Diagnosis: {diagnosis}\nVignette: {summary}"
        )

        return {
            "summary": sum_out,
            "diagnosis": diag_out,
            "question": q_out,
            "treatment": treatment_out,
            "conversation": "\n".join(history),
        }


# === Gradio UI ===
def ui_fn(user_input):
    res = simulate_interaction(user_input)
    return f"""📋 Vignette Summary:
💭 THINKING: {res['summary']['thinking']}
ANSWER: {res['summary']['output']}

🩺 Diagnosis:
💭 THINKING: {res['diagnosis']['thinking']}
ANSWER: {res['diagnosis']['output']}
T
❓ Follow-up Question:
💭 THINKING: {res['question']['thinking']}
ANSWER: {res['question']['output']}

💊 Treatment Plan:
{res['treatment']['output']}

💬 Conversation:
{res['conversation']}
"""


demo = gr.Interface(
    fn=ui_fn,
    inputs=gr.Textbox(label="Patient Response"),
    outputs=gr.Textbox(label="Doctor Simulation Output"),
    title="🧠 AI Doctor Multi-Agent Reasoning",
)

if __name__ == "__main__":
    demo.launch(share=True)