| # Primary: Qwen2.5-Coder-32B via HuggingFace (AMD hackathon Qwen prize eligible) | |
| USE_QWEN=true | |
| QWEN_API_KEY=hf_your_token_here | |
| QWEN_MODEL=Qwen/Qwen2.5-Coder-32B-Instruct | |
| QWEN_BASE_URL=https://api-inference.huggingface.co/models/Qwen/Qwen2.5-Coder-32B-Instruct/v1 | |
| # Fallback: Groq LLaMA (when Qwen unavailable) | |
| GROQ_API_KEY=your_groq_key | |
| GROQ_MODEL=llama-3.3-70b-versatile | |
| # AMD DevCloud production (vLLM on MI300X) | |
| # USE_VLLM=true | |
| # VLLM_BASE_URL=http://your-amd-cloud:8000/v1 | |
| # VLLM_MODEL=Qwen/Qwen2.5-Coder-32B-Instruct | |
| # ROCm toolchain | |
| ROCM_AVAILABLE=true | |
| HIPCC_PATH=hipcc | |
| ROCPROF_PATH=rocprof |