{"slug": "agentic-coding-harness-july-daily-driver-120gb-vram", "title": "Agentic Coding Harness - July - Daily Driver < 120GB VRAM", "summary": "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.", "body_md": "For me, I’m running ([nvidia/Qwen3.6-27B-NVFP4 · Hugging Face](https://huggingface.co/nvidia/Qwen3.6-27B-NVFP4))\n\nI found this to be VERY Stable on my DGX Spark:\n\n--host 0.0.0.0\n\n–port 8006\n\n–served-model-name qwen3.6\n\n–trust-remote-code\n\n–quantization compressed-tensors\n\n–dtype auto\n\n–kv-cache-dtype fp8\n\n–language-model-only\n\n–gpu-memory-utilization 0.80\n\n–max-model-len 131072\n\n–max-num-seqs 6\n\n–max-num-batched-tokens 16384\n\n–enable-prefix-caching\n\n–enable-chunked-prefill\n\n–enable-auto-tool-choice\n\n–tool-call-parser qwen3_xml\n\n–reasoning-parser qwen3\n\n–speculative-config ‘{“method”:“mtp”,“num_speculative_tokens”:3,“moe_backend”:“triton”}’", "url": "https://wpnews.pro/news/agentic-coding-harness-july-daily-driver-120gb-vram", "canonical_source": "https://discuss.huggingface.co/t/agentic-coding-harness-july-daily-driver-120gb-vram/177625#post_1", "published_at": "2026-07-09 16:59:25+00:00", "updated_at": "2026-07-09 17:19:02.638042+00:00", "lang": "en", "topics": ["large-language-models", "ai-infrastructure", "ai-tools", "developer-tools"], "entities": ["Nvidia", "Qwen3.6-27B-NVFP4", "DGX Spark", "Hugging Face"], "alternates": {"html": "https://wpnews.pro/news/agentic-coding-harness-july-daily-driver-120gb-vram", "markdown": "https://wpnews.pro/news/agentic-coding-harness-july-daily-driver-120gb-vram.md", "text": "https://wpnews.pro/news/agentic-coding-harness-july-daily-driver-120gb-vram.txt", "jsonld": "https://wpnews.pro/news/agentic-coding-harness-july-daily-driver-120gb-vram.jsonld"}}