Implement Custom DeepSeek Wrapper (Source: ISH_harry_potter_rag)

main.py

@@ -38,26 +38,59 @@ if os.path.exists(FAISS_PATH):
         retriever = None
 else:
     print("WARNING: FAISS index not found at path. Run ingest.py first.")
-# Hugging Face Inference API (Serverless & Free)
-if not os.getenv("HUGGINGFACEHUB_API_TOKEN"):
-    print("CRITICAL WARNING: HUGGINGFACEHUB_API_TOKEN not found. API calls will fail.")
-
-repo_id = "deepseek-ai/DeepSeek-V3.2"
-
-try:
-    print(f"Connecting to Hugging Face API ({repo_id})...")
-    llm = HuggingFaceEndpoint(
-        repo_id=repo_id,
-        task="conversational",
-        max_new_tokens=512,
-        do_sample=True,
-        temperature=0.7,
-        top_p=0.9,
-    )
-    print("Hugging Face API Client connected!")
-except Exception as e:
-    print(f"FAILED to connect to HF API: {e}")
-    llm = None
+import requests
+import json
+from langchain_core.runnables import RunnableLambda
+
+# Custom DeepSeek Connector (Cloned from Reference Repo)
+def call_deepseek_v3(prompt_input):
+    # Handle LangChain prompt objects
+    if hasattr(prompt_input, "to_string"):
+        prompt_text = prompt_input.to_string()
+    else:
+        prompt_text = str(prompt_input)
+
+    # Direct Router API used by the reference repo
+    api_url = "https://router.huggingface.co/v1/chat/completions"
+    token = os.getenv("HUGGINGFACEHUB_API_TOKEN")
+    
+    if not token:
+        return "Error: No API Token found."
+
+    headers = {
+        "Authorization": f"Bearer {token}",
+        "Content-Type": "application/json",
+    }
+    
+    payload = {
+        "model": "deepseek-ai/DeepSeek-V3.2",
+        "messages": [
+            {
+                "role": "system",
+                "content": "You are an expert on Stranger Things. Answer clearly and concisely."
+            },
+            {
+                "role": "user",
+                "content": prompt_text
+            }
+        ],
+        "temperature": 0.3,
+        "max_tokens": 512,
+        "stream": False 
+    }
+
+    try:
+        response = requests.post(api_url, headers=headers, json=payload)
+        if response.status_code != 200:
+             print(f"API Error {response.status_code}: {response.text}")
+             return f"DeepSeek Error: {response.text}"
+        return response.json()["choices"][0]["message"]["content"]
+    except Exception as e:
+        print(f"DeepSeek Connection Exception: {e}")
+        return f"Error: {e}"
+
+llm = RunnableLambda(call_deepseek_v3)
+print("DeepSeek V3.2 Client (Custom Request) initialized!")
 
 from langchain_core.prompts import PromptTemplate
 from langchain_core.output_parsers import StrOutputParser
@@ -92,7 +125,6 @@ if retriever and llm:
         {"context": retriever | format_docs, "question": RunnablePassthrough()}
         | prompt
         | llm
-        | StrOutputParser()
     )
     print("DEBUG: rag_chain constructed successfully.")
 else: