updated LLM configurations and added support for new models

backend/app/config.py

@@ -8,7 +8,7 @@ MIN_CRON_HOURS = float(os.getenv("MIN_CRON_HOURS", 0.5))
 DEFAULT_CRON_HOURS = float(os.getenv("CRON_HOURS", MIN_CRON_HOURS))
 CRON_HOURS = max(MIN_CRON_HOURS, DEFAULT_CRON_HOURS)
 SYNC_COOLDOWN_MINUTES = int(os.getenv("SYNC_COOLDOWN_MINUTES", 30))
-LLM_MODEL = os.getenv("LLM_MODEL", "qwen2:7b-instruct-q4_K_M")
+LLM_MODEL = os.getenv("LLM_MODEL", "mistral-nemo:12b")
 LLM_TIMEOUT_SECONDS = int(os.getenv("LLM_TIMEOUT_SECONDS", 180))
 OLLAMA_API_TIMEOUT_SECONDS = int(os.getenv("OLLAMA_API_TIMEOUT_SECONDS", 10))
 ARTICLE_FETCH_TIMEOUT = int(os.getenv("ARTICLE_FETCH_TIMEOUT", 30))

backend/app/services.py

@@ -206,7 +206,11 @@ class NewsFetcher:
             "prompt": prompt,
             "stream": False,
             "temperature": 0.1,
-            "format": "json"
+            "format": "json",
+            "options": {
+                "num_gpu": 1,  # Force GPU usage
+                "num_ctx": 8192,  # Context size
+            }
         }
 
         try:

backend/example.env

@@ -12,6 +12,8 @@ SYNC_COOLDOWN_MINUTES=30
 
 # LLM model to use for summarization
 LLM_MODEL=qwen2:7b-instruct-q4_K_M
+LLM_MODEL=phi3:3.8b-mini-128k-instruct-q4_0
+LLM_MODEL=mistral-nemo:12b
 
 # Timeout in seconds for LLM requests
 LLM_TIMEOUT_SECONDS=180