back to mistral

components/LLMs/Bart.py

@@ -1,50 +0,0 @@
-# components/LLMs/bart.py
-
-import os
-import requests
-from typing import Optional
-
-HF_TOKEN = os.environ.get("HF_TOKEN")
-BART_URL = "https://c5dk65n3sd14gjo1.us-east-1.aws.endpoints.huggingface.cloud"
-
-HEADERS = {
-    "Authorization": f"Bearer {HF_TOKEN}",
-    "Content-Type": "application/json"
-}
-
-
-def call_bart_summarizer(base_prompt: str, tail_prompt: str, max_length: int = 130) -> Optional[str]:
-    """
-    Calls facebook/bart-large-cnn using HF Inference API with composed prompt.
-
-    Args:
-        base_prompt (str): Instruction or high-level instruction.
-        tail_prompt (str): News content or body.
-        max_length (int): Output summary length limit.
-
-    Returns:
-        str: Cleaned summary string, or None if error.
-    """
-    full_input = f"{base_prompt.strip()}\n\n{tail_prompt.strip()}"
-    payload = {
-        "inputs": full_input,
-        "parameters": {
-            "max_length": max_length,
-            "do_sample": False
-        }
-    }
-
-    try:
-        response = requests.post(BART_URL, headers=HEADERS, json=payload, timeout=30)
-        response.raise_for_status()
-        result = response.json()
-
-        if isinstance(result, list) and result and "summary_text" in result[0]:
-            return result[0]["summary_text"].strip()
-        else:
-            print("⚠️ Unexpected BART response format:", result)
-            return None
-
-    except Exception as e:
-        print(f"⚠️ BART API call failed: {e}")
-        return None

components/LLMs/LLama3.py

@@ -1,43 +0,0 @@
-import os
-import requests
-from typing import Optional
-
-# 🔐 Environment Variables
-LLAMA3_URL = "https://c5dk65n3sd14gjo1.us-east-1.aws.endpoints.huggingface.cloud"
-HF_TOKEN = os.environ.get("HF_TOKEN")
-
-# 📜 Headers
-HEADERS = {
-    "Authorization": f"Bearer {HF_TOKEN}",
-    "Content-Type": "application/json"
-}
-
-# 🧠 Prompt builder and caller for LLaMA 3 8B (QCA)
-def call_llama3_8b(base_prompt: str, tail_prompt: str) -> Optional[str]:
-    prompt = f"<s>[INST]{base_prompt}\n\n{tail_prompt}[/INST]</s>"
-
-    try:
-        response = requests.post(
-            LLAMA3_URL,
-            headers=HEADERS,
-            json={"inputs": prompt},
-            timeout=60
-        )
-        response.raise_for_status()
-        data = response.json()
-
-        # Parse generated output
-        if isinstance(data, list) and data:
-            raw_output = data[0].get("generated_text", "")
-        elif isinstance(data, dict):
-            raw_output = data.get("generated_text", "")
-        else:
-            return None
-
-        if "[/INST]</s>" in raw_output:
-            return raw_output.split("[/INST]</s>")[-1].strip()
-        return raw_output.strip()
-
-    except Exception as e:
-        print(f"⚠️ LLaMA 3.1 8B API call failed: {e}")
-        return None

components/LLMs/TinyLLama.py

@@ -1,34 +0,0 @@
-import os
-from transformers import pipeline
-from typing import Optional
-
-# Load model just once when module is imported
-_tinyllama_pipeline = pipeline(
-    "text-generation",
-    model="TinyLlama/TinyLlama-1.1B-Chat-v1.0",
-    torch_dtype="auto",
-    device_map="auto"
-)
-
-def call_tinyllama(base_prompt: str, tail_prompt: str, max_new_tokens: int = 256) -> Optional[str]:
-    """
-    Calls TinyLlama model with an instruction-tuned prompt.
-
-    Args:
-        base_prompt (str): Instruction or system prompt.
-        tail_prompt (str): User or content-specific prompt.
-        max_new_tokens (int): Max tokens to generate.
-
-    Returns:
-        str or None: The generated summary content.
-    """
-    prompt = f"<s>[INST]{base_prompt}\n\n{tail_prompt}[/INST]</s>"
-    try:
-        result = _tinyllama_pipeline(prompt, max_new_tokens=max_new_tokens)
-        output = result[0]["generated_text"]
-        if "[/INST]" in output:
-            return output.split("[/INST]")[-1].strip()
-        return output.strip()
-    except Exception as e:
-        print(f"⚠️ TinyLlama error: {e}")
-        return None

components/generators/daily_feed.py

@@ -9,9 +9,6 @@ from llama_index.core.query_engine import RetrieverQueryEngine
 from llama_index.core.schema import Document
 from llama_index.core.settings import Settings
 from components.LLMs.Mistral import call_mistral
-from components.LLMs.TinyLLama import call_tinyllama
-from components.LLMs.Bart import call_bart_summarizer
-from components.LLMs.LLama3 import call_llama3_8b
 
 # ✅ Disable implicit LLM usage
 Settings.llm = None
@@ -27,6 +24,7 @@ redis_client = redis.Redis.from_url(REDIS_URL, decode_responses=True)
 TOPICS = ["India news", "World news", "Tech news", "Finance news", "Sports news"]
 
 # 🧠 Base summarization prompt (used for all topics)
+# 🧠 Define the base summarization prompt
 BASE_PROMPT = (
     "You are Nuse’s official news summarizer — insightful, punchy, and always on point.\n"
     "Your job is to scan the content below and extract the key news items. For each item, craft a crisp summary (15–20 words). Avoid using any emojis.\n"
@@ -35,7 +33,7 @@ BASE_PROMPT = (
     "Example format:\n"
     "- India stuns Australia in a last-ball thriller at the World Cup finals\n"
     "- U.S. imposes sweeping tariffs on Chinese tech giants, rattling global markets\n"
-    "- Ceasefire breakthrough: Netanyahu (Prime minister of Isreal) bows to pressure after week-long escalation\n"
+    "- Ceasefire breakthrough: Netanyahu (Prime minister of Israel) bows to pressure after week-long escalation\n"
     "\n"
     "If you are mentioning a person, include their designation in brackets. For example: Jeff Bezos (Amazon CEO), Narendra Modi (Prime minister of India).\n"
     "If you're referencing a post like 'NATO Chief', also include the name of the person who holds the post.\n"
@@ -45,13 +43,12 @@ BASE_PROMPT = (
     "Return only the summary block — no extra commentary, no prompt repetition."
 )
 
-# ✂️ Summarize top N documents
 def summarize_topic(docs: List[str], topic: str) -> List[Dict]:
     feed = []
     for doc in docs[:5]:
         tail_prompt = f"Topic: {topic}\n\n{doc.strip()}"
-        print(f"\n📤 Prompt tail for Mistral:\n{tail_prompt[:300]}...\n")
-        summary_block = call_llama3_8b(base_prompt=BASE_PROMPT, tail_prompt=tail_prompt)
+        print(f"\n📤 Prompt tail for LLaMA-3 8B:\n{tail_prompt[:300]}...\n")
+        summary_block = call_mistral(base_prompt=BASE_PROMPT, tail_prompt=tail_prompt)
 
         if summary_block:
             for line in summary_block.splitlines():
@@ -62,10 +59,11 @@ def summarize_topic(docs: List[str], topic: str) -> List[Dict]:
                         feed.append({
                             "summary": clean_summary,
                             "image_url": "https://source.unsplash.com/800x600/?news",
-                            "article_link": "https://google.com/search?q=" + topic.replace(" ", "+")
+                            "article_link": f"https://google.com/search?q={topic.replace(' ', '+')}"
                         })
     return feed
 
+
 # ⚡ Generate and cache daily feed
 def generate_and_cache_daily_feed(documents: List[Document]):
     index = VectorStoreIndex.from_documents(documents)