no message

main.py

@@ -1,24 +1,25 @@
-from fastapi import FastAPI
+from fastapi import FastAPI, HTTPException
 from fastapi.responses import StreamingResponse
+from fastapi.responses import JSONResponse
 from pydantic import BaseModel
 from huggingface_hub import InferenceClient
 import uvicorn
 from typing import Generator
-import json
+import json  # Asegúrate de que esta línea esté al principio del archivo
 import nltk
 import os
 from transformers import pipeline
 
-# Set up the environment for NLTK
+
 nltk.data.path.append(os.getenv('NLTK_DATA'))
 
-# Initialize the FastAPI app
 app = FastAPI()
 
 # Initialize the InferenceClient with your model
 client = InferenceClient("mistralai/Mistral-7B-Instruct-v0.2")
 
-# Initialize the summarization pipeline
+# summarizer = pipeline("summarization", model="sshleifer/distilbart-cnn-12-6")
+
 summarizer = pipeline("summarization", model="sshleifer/distilbart-cnn-12-6")
 
 class Item(BaseModel):
@@ -30,24 +31,23 @@ class Item(BaseModel):
     top_p: float = 0.15
     repetition_penalty: float = 1.0
 
-def summarize_history(history):
-    # Concatenate all history entries into a single string
-    full_history = " ".join(entry['content'] for entry in history if entry['role'] == 'user')
-    # Summarize the history
-    summarized_history = summarizer(full_history, max_length=1024, truncation=True)
-    return summarized_history[0]['summary_text']
-
 def format_prompt(current_prompt, history):
     formatted_history = "<s>"
-    formatted_history += f"[HISTORY] {history} [/HISTORY]"
+    for entry in history:
+        if entry["role"] == "user":
+            formatted_history += f"[USER] {entry['content']} [/USER]"
+        elif entry["role"] == "assistant":
+            formatted_history += f"[ASSISTANT] {entry['content']} [/ASSISTANT]"
     formatted_history += f"[USER] {current_prompt} [/USER]</s>"
     return formatted_history
 
+
 def generate_stream(item: Item) -> Generator[bytes, None, None]:
-    summarized_history = summarize_history(item.history)
-    formatted_prompt = format_prompt(item.prompt, summarized_history)
-    input_token_count = len(nltk.word_tokenize(formatted_prompt))
+    formatted_prompt = format_prompt(f"{item.system_prompt}, {item.prompt}", item.history)
+    # Estimate token count for the formatted_prompt
+    input_token_count = len(nltk.word_tokenize(formatted_prompt))  # NLTK tokenization
 
+    # Ensure total token count doesn't exceed the maximum limit
     max_tokens_allowed = 32768
     max_new_tokens_adjusted = max(1, min(item.max_new_tokens, max_tokens_allowed - input_token_count))
 
@@ -60,16 +60,33 @@ def generate_stream(item: Item) -> Generator[bytes, None, None]:
         "seed": 42,
     }
 
+    # Stream the response from the InferenceClient
     for response in client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True):
+        # This assumes 'details=True' gives you a structure where you can access the text like this
         chunk = {
             "text": response.token.text,
-            "complete": response.generated_text is not None
+            "complete": response.generated_text is not None  # Adjust based on how you detect completion
         }
         yield json.dumps(chunk).encode("utf-8") + b"\n"
 
+
+class SummarizeRequest(BaseModel):
+    text: str
+
 @app.post("/generate/")
 async def generate_text(item: Item):
+    # Stream response back to the client
     return StreamingResponse(generate_stream(item), media_type="application/x-ndjson")
 
+@app.post("/summarize")
+async def summarize_text(request: SummarizeRequest):
+    try:
+        # Perform the summarization
+        summary = summarizer(request.text, max_length=130, min_length=30, do_sample=False)
+        return JSONResponse(content={"summary": summary[0]['summary_text']})
+    except Exception as e:
+        # Handle exceptions that could arise during summarization
+        raise HTTPException(status_code=500, detail=str(e))
+    
 if __name__ == "__main__":
     uvicorn.run(app, host="0.0.0.0", port=8000)