switch to fast API and mistral

app.py

@@ -1,74 +1,54 @@
-#refer llama recipes for more info https://github.com/huggingface/huggingface-llama-recipes/blob/main/inference-api.ipynb
-#huggingface-llama-recipes : https://github.com/huggingface/huggingface-llama-recipes/tree/main
-import gradio as gr
-from openai import OpenAI
-import os
-
-css = '''
-.gradio-container{max-width: 890px !important}
-h1{text-align:center}
-footer {
-    visibility: hidden
-}
-'''
-
-ACCESS_TOKEN = os.getenv("HF_TOKEN")
-
-client = OpenAI(
-    base_url="https://api-inference.huggingface.co/v1/",
-    api_key=ACCESS_TOKEN,
-)
-
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
-
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
-
-    messages.append({"role": "user", "content": message})
-
-    response = ""
-    
-    for message in  client.chat.completions.create(
-        model="meta-llama/Meta-Llama-3.1-70B-Instruct",
-        max_tokens=max_tokens,
-        stream=True,
+# from https://huggingface.co/spaces/iiced/mixtral-46.7b-fastapi/blob/main/main.py
+from fastapi import FastAPI
+from pydantic import BaseModel
+from huggingface_hub import InferenceClient
+import uvicorn
+
+
+app = FastAPI()
+
+client = InferenceClient("mistralai/Mistral-7B-Instruct-v0.3")
+
+class Item(BaseModel):
+    prompt: str
+    history: list
+    system_prompt: str
+    temperature: float = 0.0
+    max_new_tokens: int = 1048
+    top_p: float = 0.15
+    repetition_penalty: float = 1.0
+
+def format_prompt(message, history):
+    prompt = "<s>"
+    for user_prompt, bot_response in history:
+        prompt += f"[INST] {user_prompt} [/INST]"
+        prompt += f" {bot_response}</s> "
+    prompt += f"[INST] {message} [/INST]"
+    return prompt
+
+def generate(item: Item):
+    temperature = float(item.temperature)
+    if temperature < 1e-2:
+        temperature = 1e-2
+    top_p = float(item.top_p)
+
+    generate_kwargs = dict(
         temperature=temperature,
+        max_new_tokens=item.max_new_tokens,
         top_p=top_p,
-        messages=messages,
-    ):
-        token = message.choices[0].delta.content
-        
-        response += token
-        yield response
-
-demo = gr.ChatInterface(
-    respond,
-    additional_inputs=[
-        gr.Textbox(value="", 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",
-        ),
-        
-    ],
-    css=css,
-    theme="allenai/gradio-theme",
-)
-if __name__ == "__main__":
-    demo.launch()
+        repetition_penalty=item.repetition_penalty,
+        do_sample=True,
+        seed=42,
+    )
+
+    formatted_prompt = format_prompt(f"{item.system_prompt}, {item.prompt}", item.history)
+    stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
+    output = ""
+
+    for response in stream:
+        output += response.token.text
+    return output
+
+@app.post("/generate/")
+async def generate_text(item: Item):
+    return {"response": generate(item)}