engine = "vllm" name = "Qwen-AgentWorld-35B-A3B (vLLM)" docker_image = "ghcr.io/timothystewart6/vllm-gb10:v0.23.0-gb10.1" model = "Qwen/Qwen-AgentWorld-35B-A3B" required_nodes = 2 tensor_parallel_size = 2 pipeline_parallel_size = 1 distributed_backend = "mp" # 35B total / 3B active MoE — world model for agentic environment simulation. # 2× DGX Spark (GB10), 128 GB shared each. # TP=2 → ~35 GB/node for weights, room for 262K context. # Requires --language-model-only (no vision weights in checkpoint). entrypoint = ["vllm", "serve"] args = """ --port 30000 --max-model-len 262144 --reasoning-parser qwen3 --language-model-only --trust-remote-code --gpu-memory-utilization 0.90 --enable-prefix-caching """