glm53-kindling-2x: second perf pass (post-campaign cooldown) — d0 c1 51.4 vs 39.6 hot: run-variance baseline

recipes/glm53-kindling-2x/reports/2026-10-15_112928_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": "2026-10-15T11-34-50.926324Z_150314ac", "status": "completed", "report_path": "/tmp/sm121-bench-2026-10-15_112928_perf/runs/2026/10/2026-10-15T11-34-50.926324Z_150314ac.md"}

recipes/glm53-kindling-2x/reports/2026-10-15_112928_perf.md

@@ -0,0 +1,36 @@
+# Throughput Benchmark — glm53
+
+- **Run ID**: `2026-10-15T11-34-50.926324Z_150314ac`
+- **Date**: `2026-10-15T11:34:50.926390+00:00`
+- **Mode**: throughput-only
+- **tool-eval-bench**: `v2.8.0`
+- **Backend**: vllm
+- **Server**: http://***:30000/v1
+- **Model (API)**: `glm53`
+- **Model (Root)**: `nvidia/GLM-5.3-Flash-NVFP4`
+
+## Inference Engine
+
+| Property | Value |
+|---|---|
+| Engine | vLLM 0.31.0 |
+| Max Model Length | 524,288 |
+| Quantization | NVFP4 |
+| 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 | 2,303 | 51.4 | 1,013 | 3,484 | 2048+128 |
+| pp2048 tg128 @ d0 c2 | 1,687 | 48.3 | 3,501 | 7,864 | 2048+127.8 |
+| ctx pp4096 tg128 @ d4096 | 2,746 | 43.8 | 1,616 | 4,518 | 2048+128 |
+| pp2048 tg128 @ d4096 | 613 | 51.7 | 5,810 | 8,266 | 2048+128 |
+| ctx pp4096 tg128 @ d4096 c2 | 2,591 | 59.5 | 3,172 | 7,346 | 2048+128 |
+| pp2048 tg128 @ d4096 c2 | 699 | 47.1 | 5,404 | 10,165 | 2048+128 |
+| ctx pp8192 tg128 @ d8192 | 2,406 | 45.0 | 3,625 | 6,446 | 2048+128 |
+| pp2048 tg128 @ d8192 | 865 | 56.2 | 2,486 | 4,746 | 2048+128 |
+| ctx pp8192 tg128 @ d8192 c2 | 2,037 | 35.2 | 6,556 | 11,995 | 2048+127.9 |
+| pp2048 tg128 @ d8192 c2 | 761 | 36.5 | 3,930 | 8,789 | 2048+127.9 |