Update main.py

main.py

@@ -1,19 +1,20 @@
 from fastapi import FastAPI
 from pydantic import BaseModel
-from huggingface_hub import InferenceClient
+from llama_cpp import Llama
 import uvicorn
 import prompt_style
 import os
 
+model_id = "failspy/Meta-Llama-3-8B-Instruct-abliterated-v3-GGUF"
+model = Llama.from_pretrained(repo_id=model_id, filename="*-v3_q6.gguf", n_gpu_layers=-1, n_ctx=4096, verbose=False)
 
-model_id = "failspy/Meta-Llama-3-8B-Instruct-abliterated-v3"
-client = InferenceClient(token=os.getenv('HF_TOKEN'), model=model_id)
+# model_id = "failspy/Meta-Llama-3-8B-Instruct-abliterated-v3"
+# client = InferenceClient(token=os.getenv('HF_TOKEN'), model=model_id)
 
 class Item(BaseModel):
     prompt: str
     history: list
     system_prompt: str
-    token:str
     temperature: float = 0.6
     max_new_tokens: int = 1024
     top_p: float = 0.95
@@ -26,33 +27,21 @@ def format_prompt(item: Item):
     messages = [
         {"role": "system", "content": prompt_style.data},
     ]
-    for it in item.history:
-        messages.append[{"role" : "user", "content": it[0]}]
-        messages.append[{"role" : "assistant", "content": it[1]}]
+    for it in history:
+        messages.append({"role" : "user", "content": it[0]})
+        messages.append({"role" : "assistant", "content": it[1]})
+    messages.append({"role" : "user", "content": item.prompt})
     return messages
 
 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,
-        repetition_penalty=item.repetition_penalty,
-        do_sample=True,
-        seed=item.seed,
-    )
-
     formatted_prompt = format_prompt(item)
-    stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
-    output = ""
+    output = model.create_chat_completion(messages=formatted_prompt, seed=item.seed, 
+                                          temperature=item.temperature,
+                                          max_tokens=item.max_new_tokens)
+
 
-    for response in stream:
-        output += response.token.text
-    return output
+    out = output['choices'][0]['message']['content']
+    return out
 
 @app.post("/generate/")
 async def generate_text(item: Item):