Update main.py

main.py

@@ -6,53 +6,52 @@ import uvicorn
 
 app = FastAPI()
 
-client = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")
+
+client = InferenceClient("deepseek-ai/deepseek-llm-67b-chat")
 
 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):
-    print("````")
-    print(message)
-    print("++++")
-    print(history)
-    print("````")
-    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]"
+    temperature: float = 0.7  
+    max_new_tokens: int = 1024
+    top_p: float = 0.9
+    repetition_penalty: float = 1.1
+
+def format_prompt(message, history, system_prompt):
+    prompt = f"<｜begin▁of▁sentence｜>{system_prompt}\n\n" if system_prompt else ""
+    
+    for user_msg, bot_res in history:
+        prompt += f"User: {user_msg}\n\nAssistant: {bot_res}\n\n"
+        
+    prompt += f"User: {message}\n\nAssistant: "
     return prompt
 
 async def generate_stream(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,
-        repetition_penalty=item.repetition_penalty,
-        do_sample=True,
-        seed=42,
+    generate_kwargs = {
+        "temperature": max(item.temperature, 0.01),
+        "max_new_tokens": item.max_new_tokens,
+        "top_p": item.top_p,
+        "repetition_penalty": item.repetition_penalty,
+        "do_sample": True,
+        "seed": 42,
+    }
+
+
+    formatted_prompt = format_prompt(
+        item.prompt, 
+        item.history,
+        item.system_prompt
+    )
+    
+    stream = client.text_generation(
+        formatted_prompt,
+        stream=True,
+        **generate_kwargs
     )
-
-    formatted_prompt = format_prompt(f"{item.system_prompt} [/INST] Ok..! </s> [INST] {item.prompt}", item.history)
-    print(formatted_prompt)
-    print("=======")
-    print(item.history)
-    stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
 
     for response in stream:
-        yield response.token.text  # Stream each token as it's received
+        yield response  
 
 @app.post("/generate/")
 async def generate_text(item: Item):