new model

app.py

@@ -41,8 +41,8 @@ from pydantic.v1 import BaseModel, Field
 from huggingface_hub import snapshot_download
 
 # Local transformers pipeline (no API token required)
-from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline
-from langchain_community.llms import HuggingFacePipeline
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline  # (still imported; not used in Hub mode)
+from langchain_community.llms import HuggingFacePipeline, HuggingFaceHub  # <-- ADDED: HuggingFaceHub import
 
 # Theming + URL list
 import theme
@@ -170,6 +170,8 @@ SYSTEM_TEMPLATE = (
     "{format_instructions}"
 )
 
+# NOTE: Your original pattern kept; if you prefer, you can also do:
+# ChatPromptTemplate.from_template(SYSTEM_TEMPLATE).partial(format_instructions=parser.get_format_instructions())
 qa_prompt = ChatPromptTemplate.from_template(
     SYSTEM_TEMPLATE,
     partial_variables={"format_instructions": parser.get_format_instructions()},
@@ -177,14 +179,22 @@ qa_prompt = ChatPromptTemplate.from_template(
 
 
 # =============================
-# 4) LLM — local, token-free
+# 4) LLM — HuggingFace Hub (Mixtral)
 # =============================
-llm = HuggingFaceHub( repo_id="mistralai/Mixtral-8x7B-v0.1", 
-                     task="text-generation", 
-                     model_kwargs={ "max_new_tokens": 2000, 
-                                   "top_k": 30, 
-                                   "temperature": 0.1, 
-                                   "repetition_penalty": 1.03 }, )
+# REQUIREMENT: set env var HUGGINGFACEHUB_API_TOKEN
+# (Settings → Variables & secrets in your Space)
+llm = HuggingFaceHub(
+    repo_id="mistralai/Mixtral-8x7B-v0.1",
+    task="text-generation",
+    model_kwargs={
+        "max_new_tokens": 2000,
+        "top_k": 30,
+        "temperature": 0.1,
+        "repetition_penalty": 1.03,
+        # You may also pass: "return_full_text": False
+    },
+)
+
 
 # ===========================================
 # 5) Chain (memory + robust JSON extraction)

requirements.txt

@@ -6,12 +6,14 @@ sentence-transformers>=3.0.0
 tensorflow==2.17.0
 typing-extensions>=4.7
 tf-keras==2.17.0
+pillow
 
 # LangChain split packages (v0.2 style)
 langchain==0.2.12
 langchain-community==0.2.10
 langchain-text-splitters==0.2.2
 langchain-core==0.2.27
+langchain-huggingface>=0.0.3
 
 # Vector store
 chromadb==0.5.3