Spaces:
Runtime error
Runtime error
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,98 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import logging
|
3 |
+
import gradio as gr
|
4 |
+
from PIL import Image
|
5 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
6 |
+
from llama_cpp import Llama
|
7 |
+
from huggingface_hub import hf_hub_download
|
8 |
+
|
9 |
+
logging.basicConfig(level=logging.INFO)
|
10 |
+
logger = logging.getLogger(__name__)
|
11 |
+
|
12 |
+
# ------------------------------
|
13 |
+
# πΉ Load Bioformer-8L Model
|
14 |
+
# ------------------------------
|
15 |
+
BIOFORMER_MODEL = "bioformers/bioformer-8L"
|
16 |
+
bioformer_tokenizer = AutoTokenizer.from_pretrained(BIOFORMER_MODEL)
|
17 |
+
bioformer_model = AutoModelForCausalLM.from_pretrained(BIOFORMER_MODEL)
|
18 |
+
|
19 |
+
# ------------------------------
|
20 |
+
# πΉ Load DeepSeek-R1-Distill-Qwen-7B-GGUF Model
|
21 |
+
# ------------------------------
|
22 |
+
DEEPSEEK_REPO = "lmstudio-community/DeepSeek-R1-Distill-Qwen-7B-GGUF"
|
23 |
+
DEEPSEEK_FILENAME = "DeepSeek-R1-Distill-Qwen-7B-Q4_0.gguf"
|
24 |
+
|
25 |
+
model_path = hf_hub_download(repo_id=DEEPSEEK_REPO, filename=DEEPSEEK_FILENAME)
|
26 |
+
|
27 |
+
llm = Llama(
|
28 |
+
model_path=model_path,
|
29 |
+
n_ctx=4096,
|
30 |
+
n_gpu_layers=0, # CPU inference
|
31 |
+
logits_all=True,
|
32 |
+
n_batch=256
|
33 |
+
)
|
34 |
+
|
35 |
+
logger.info("Models Loaded Successfully.")
|
36 |
+
|
37 |
+
# ------------------------------
|
38 |
+
# πΉ Unified Medical Prompt
|
39 |
+
# ------------------------------
|
40 |
+
UNIFIED_MEDICAL_PROMPT = """
|
41 |
+
You are an advanced Medical AI Assistant capable of providing thorough,
|
42 |
+
comprehensive answers for a wide range of medical specialties:
|
43 |
+
General Practice, Radiology, Cardiology, Neurology, Psychiatry, Pediatrics,
|
44 |
+
Endocrinology, Oncology, and more.
|
45 |
+
|
46 |
+
You can:
|
47 |
+
1) Analyze images if provided (Radiology).
|
48 |
+
2) Retrieve relevant documents from a knowledge base (Vector Store).
|
49 |
+
3) Provide scientific, evidence-based explanations and references when possible.
|
50 |
+
|
51 |
+
Always strive to provide a detailed, helpful, and empathetic response.
|
52 |
+
"""
|
53 |
+
|
54 |
+
# ------------------------------
|
55 |
+
# πΉ Chat Function
|
56 |
+
# ------------------------------
|
57 |
+
def chat_with_agent(user_query, image_file=None):
|
58 |
+
# Combine context
|
59 |
+
combined_context = f"""
|
60 |
+
{UNIFIED_MEDICAL_PROMPT}
|
61 |
+
|
62 |
+
Patient Query: "{user_query}"
|
63 |
+
Your Response:
|
64 |
+
"""
|
65 |
+
|
66 |
+
# Generate response using DeepSeek-R1-Distill model
|
67 |
+
response_accumulator = ""
|
68 |
+
for token in llm(
|
69 |
+
prompt=combined_context,
|
70 |
+
max_tokens=1024,
|
71 |
+
temperature=0.7,
|
72 |
+
top_p=0.9,
|
73 |
+
stream=True
|
74 |
+
):
|
75 |
+
partial_text = token["choices"][0]["text"]
|
76 |
+
response_accumulator += partial_text
|
77 |
+
yield response_accumulator
|
78 |
+
|
79 |
+
# ------------------------------
|
80 |
+
# πΉ Gradio Interface
|
81 |
+
# ------------------------------
|
82 |
+
with gr.Blocks(title="π₯ Llama3-Med AI Assistant") as demo:
|
83 |
+
gr.Markdown("""
|
84 |
+
# π₯ Llama3-Med AI Assistant
|
85 |
+
_Your intelligent medical assistant powered by advanced AI._
|
86 |
+
""")
|
87 |
+
|
88 |
+
with gr.Row():
|
89 |
+
user_input = gr.Textbox(label="π¬ Ask a medical question", placeholder="Type your question here...")
|
90 |
+
image_file = gr.Image(label="π· Upload Medical Image (Optional)", type="filepath")
|
91 |
+
|
92 |
+
submit_btn = gr.Button("π Submit", variant="primary")
|
93 |
+
output_text = gr.Textbox(label="π Assistant's Response", interactive=False, lines=25)
|
94 |
+
|
95 |
+
submit_btn.click(fn=chat_with_agent, inputs=[user_input, image_file], outputs=output_text)
|
96 |
+
|
97 |
+
if __name__ == "__main__":
|
98 |
+
demo.queue().launch(server_name="0.0.0.0", server_port=7860)
|