diff --git a/ai-weather-agent/Procfile b/ai-weather-agent/Procfile index 7ba3ee7..f5008b0 100644 --- a/ai-weather-agent/Procfile +++ b/ai-weather-agent/Procfile @@ -1,4 +1,4 @@ -max-serve-llm: (magic global install max-pipelines && magic global update max-pipelines) || true; MAX_SERVE_PORT=8000 MAX_SERVE_HOST=localhost HUGGING_FACE_HUB_TOKEN=$(cat backend/.env | grep HUGGING_FACE_HUB_TOKEN | cut -d '=' -f2) max-pipelines serve --huggingface-repo-id=modularai/Llama-3.1-8B-Instruct-GGUF --max-length 2048 -max-serve-embedding: (magic global install max-pipelines && magic global update max-pipelines) || true; MAX_SERVE_PORT=7999 MAX_SERVE_HOST=localhost HUGGING_FACE_HUB_TOKEN=$(cat backend/.env | grep HUGGING_FACE_HUB_TOKEN | cut -d '=' -f2) max-pipelines serve --huggingface-repo-id=sentence-transformers/all-mpnet-base-v2 +max-serve-llm: (magic global install max-pipelines && magic global update max-pipelines) || true; MAX_SERVE_PORT=8010 MAX_SERVE_HOST=0.0.0.0 HUGGING_FACE_HUB_TOKEN=$(cat backend/.env | grep HUGGING_FACE_HUB_TOKEN | cut -d '=' -f2) max-pipelines serve --huggingface-repo-id=modularai/Llama-3.1-8B-Instruct-GGUF --max-length 2048 +max-serve-embedding: (magic global install max-pipelines && magic global update max-pipelines) || true; MAX_SERVE_PORT=7999 MAX_SERVE_HOST=0.0.0.0 HUGGING_FACE_HUB_TOKEN=$(cat backend/.env | grep HUGGING_FACE_HUB_TOKEN | cut -d '=' -f2) max-pipelines serve --huggingface-repo-id=sentence-transformers/all-mpnet-base-v2 backend: cd backend && magic run backend frontend: cd frontend && magic run frontend diff --git a/ai-weather-agent/backend/src/main.py b/ai-weather-agent/backend/src/main.py index 3dca0d1..311bdab 100644 --- a/ai-weather-agent/backend/src/main.py +++ b/ai-weather-agent/backend/src/main.py @@ -76,7 +76,7 @@ async def lifespan(app: FastAPI): CACHE_TTL = 3600 # 1 hour -LLM_SERVER_URL = os.getenv("LLM_SERVER_URL", "http://0.0.0.0:8000/v1") +LLM_SERVER_URL = os.getenv("LLM_SERVER_URL", "http://0.0.0.0:8010/v1") EMBEDDING_SERVER_URL = os.getenv("EMBEDDING_SERVER_URL", "http://0.0.0.0:7999/v1") EMBEDDING_MODEL = os.getenv( "EMBEDDING_MODEL", "sentence-transformers/all-mpnet-base-v2"