glm53-fp4local-2x perf: d0 pp 1830/tg 38.9, d8192 c1 tg 34.9 (c2 42.2), TTFT 1.2s — official NVFP4 QAD decode ~35-40 t/s vs kindling 2x 52-59 t/s

recipes/glm53-flash-fp4-local-2x/reports/2026-10-17_184045_perf.log

@@ -0,0 +1,3 @@
+Pseudo-terminal will not be allocated because stdin is not a terminal.
+{"event": "backend_detected", "backend": "vllm"}
+{"event": "run_saved", "run_type": "perf", "run_id": "2026-10-17T18-45-41.609973Z_42bca832", "status": "completed", "report_path": "/tmp/sm121-bench-2026-10-17_184045_perf/runs/2026/10/2026-10-17T18-45-41.609973Z_42bca832.md"}

recipes/glm53-flash-fp4-local-2x/reports/2026-10-17_184045_perf.md

@@ -0,0 +1,35 @@
+# Throughput Benchmark — glm53-fp4-local
+
+- **Run ID**: `2026-10-17T18-45-41.609973Z_42bca832`
+- **Date**: `2026-10-17T18:45:41.610033+00:00`
+- **Mode**: throughput-only
+- **tool-eval-bench**: `v2.8.0`
+- **Backend**: vllm
+- **Server**: http://***:30000/v1
+- **Model (API)**: `glm53-fp4-local`
+- **Model (Root)**: `nvidia/GLM-5.3-Flash-FP4-Local`
+
+## Inference Engine
+
+| Property | Value |
+|---|---|
+| Engine | vLLM 0.1.dev21830+g8244d1c44.d20261015 |
+| Max Model Length | 500,000 |
+| Host | `pluto` |
+| Platform | `Linux-7.2.9-arch1-1-x86_64-with-glibc2.44` |
+| Python | 3.12.15 |
+
+## Results
+
+| Test | pp t/s | tg t/s | TTFT (ms) | Total (ms) | Tokens |
+|---|---:|---:|---:|---:|---:|
+| pp2048 tg128 @ d0 | 1,830 | 38.9 | 1,197 | 4,465 | 2048+128 |
+| pp2048 tg128 @ d0 c2 | 1,664 | 39.8 | 1,838 | 6,895 | 2048+128 |
+| ctx pp4096 tg128 @ d4096 | 1,945 | 32.3 | 2,185 | 6,121 | 2048+128 |
+| pp2048 tg128 @ d4096 | 1,687 | 37.0 | 1,286 | 4,718 | 2048+128 |
+| ctx pp4096 tg128 @ d4096 c2 | 1,886 | 46.0 | 4,225 | 9,402 | 2048+128 |
+| pp2048 tg128 @ d4096 c2 | 1,556 | 40.7 | 1,963 | 7,376 | 2048+128 |
+| ctx pp8192 tg128 @ d8192 | 2,030 | 32.3 | 4,114 | 8,042 | 2048+128 |
+| pp2048 tg128 @ d8192 | 1,653 | 34.9 | 1,311 | 4,953 | 2048+128 |
+| ctx pp8192 tg128 @ d8192 c2 | 2,013 | 47.2 | 8,017 | 13,158 | 2048+128 |
+| pp2048 tg128 @ d8192 c2 | 1,531 | 42.2 | 2,001 | 7,079 | 2048+128 |