.env.production

# ============================================================
# OncoAgent — PRODUCTION Environment (AMD Instinct MI300X)
# Copy to .env on the GPU droplet before deploying.
# ============================================================

# Set via: export HF_TOKEN=your_token_here
HF_TOKEN=

# --- Hardware ---
ROCM_PATH=/opt/rocm
DEVICE=cuda
HSA_OVERRIDE_GFX_VERSION=9.4.2
TENSOR_PARALLEL_SIZE=1

# --- Model Tier IDs ---
TIER1_MODEL_ID=Qwen/Qwen3.5-9B
TIER2_MODEL_ID=Qwen/Qwen3.6-27B
BASE_MODEL_ID=Qwen/Qwen3.5-9B

# --- Inference Backend (Local vLLM) ---
VLLM_API_BASE=http://localhost:8000/v1
VLLM_API_KEY=EMPTY

# --- Local LoRA Adapters (MI300X Optimized) ---
USE_LOCAL_ADAPTERS=false
LOCAL_ADAPTER_PATH=models/oncoagent_adapters/tier1/checkpoint-1000/

# --- Logging ---
LOG_LEVEL=INFO