Agentic Coding Harness - July - Daily Driver < 120GB VRAM

A developer reports stable performance running Nvidia's Qwen3.6-27B-NVFP4 model on a DGX Spark system with under 120GB VRAM, sharing specific configuration parameters including quantization, KV-cache dtype, and speculative decoding settings.

For me, I’m running nvidia/Qwen3.6-27B-NVFP4 · Hugging Face https://huggingface.co/nvidia/Qwen3.6-27B-NVFP4 I found this to be VERY Stable on my DGX Spark: --host 0.0.0.0 –port 8006 –served-model-name qwen3.6 –trust-remote-code –quantization compressed-tensors –dtype auto –kv-cache-dtype fp8 –language-model-only –gpu-memory-utilization 0.80 –max-model-len 131072 –max-num-seqs 6 –max-num-batched-tokens 16384 –enable-prefix-caching –enable-chunked-prefill –enable-auto-tool-choice –tool-call-parser qwen3 xml –reasoning-parser qwen3 –speculative-config ‘{“method”:“mtp”,“num speculative tokens”:3,“moe backend”:“triton”}’