diff --git a/.github/COPYCATS.md b/.github/COPYCATS.md index 83b9b18..7ca924e 100644 --- a/.github/COPYCATS.md +++ b/.github/COPYCATS.md @@ -46,6 +46,7 @@ The machine-readable log is [`copycats.json`](./copycats.json) — one entry per | 2026-06-25 | #9 | `glorysr1209-png` | #6 (`galuis116`) | gguf metadata desync; 7/8 added lines identical. Account already blocked as sybil. | | 2026-06-25 | #54 | `kiannidev` | #53 (`James-CUDA`) | maintainer-flagged: same decode change as #53. Below auto threshold at the time; strike 1. **2-day penalty** (first-time contributor leniency). | | 2026-07-10 | #326 | `inference2026` | #195 (`fansilas`) | **false positive (cleared).** Per-function template-launch line matched at 100%; PR-level overlap only 27%. `blocked: false`, `penalty_days: 0`. | +| 2026-07-10 | #338 | `inference2026` | #318 (`Paral1995`) | **false positive (cleared).** 3-line GQA-4 `g2` build fix; 100% literal overlap with #318 call-site boilerplate. Same diff as #336 (merged first). `blocked: false`, `penalty_days: 0`. | > `glorysr1209-png` also opened #13 and #15 (same bug-clusters as #11 / #12) but with different > code — not literal copies, so not logged here; they were closed under the sybil block instead. diff --git a/.github/FLAGGED.md b/.github/FLAGGED.md index 87e192c..a0402f4 100644 --- a/.github/FLAGGED.md +++ b/.github/FLAGGED.md @@ -126,3 +126,10 @@ containment matched a 7-token `fa_split_gqa_kernel<...>` template instantiation (100% boilerplate overlap); actual PR-level containment vs #195 was 27%. Maintainer cleared: removed from `blocked-contributors.txt`, label `copycat-cleared`, #326 reopened for eval. + +#338 was auto-blocked as ≥85% copycat of #318 — **false positive**: 3-line build fix +routing the GQA-4 hd256 branch through `fa_launch_combine_dispatch_hd256` after #300 +removed `g2`; added lines matched #318's existing call-site pattern at 100% (tiny-PR +literal rule). Independent convergent fix with #336 (merged ~14 min earlier). +Maintainer cleared: label `copycat-cleared`, #338 reopened. Account was not on +`blocked-contributors.txt`. diff --git a/.github/copycats.json b/.github/copycats.json index 626793b..2168f12 100644 --- a/.github/copycats.json +++ b/.github/copycats.json @@ -66,5 +66,16 @@ "strike": 0, "containment": 0.27, "note": "false positive cleared: per-function template-launch matched #195 at 100%; PR-level overlap 27%" + }, + { + "pr": 338, + "author": "inference2026", + "original": 318, + "date": "2026-07-10", + "blocked": false, + "penalty_days": 0, + "strike": 0, + "containment": 1.0, + "note": "false positive cleared: 3-line GQA-4 g2 build fix; 100% literal overlap with #318 call-site boilerplate; same diff as #336 (merged first)" } ] \ No newline at end of file diff --git a/dashboard/data.js b/dashboard/data.js index 6bb38df..0330439 100644 --- a/dashboard/data.js +++ b/dashboard/data.js @@ -137,6 +137,438 @@ window.SPARKINFER = { } ], "prs": [ + { + "num": 331, + "title": "perf(moe): complete gate_up->quant_h->down PDL chain for bs=1 decode", + "areas": [ + "kernels" + ], + "label": "S", + "tps": 365.46, + "delta_pct": 3.5, + "top1": 0.9665, + "kl": 0.0203, + "url": "https://github.com/gittensor-ai-lab/sparkinfer/pull/331", + "model": "bidir", + "eval_mode": "longctx", + "score_context": 32768, + "best_context_label": "32k-context", + "context_gains_pct": { + "128-context": 0.32, + "512-context": -0.1, + "4k-context": -0.11, + "16k-context": 1.81, + "32k-context": 3.52 + }, + "regression_labels": [], + "mode": "bidir", + "label_qwen35": "XS", + "label_qwen36": "S", + "pass_qwen35": true, + "pass_qwen36": true, + "score_qwen35": { + "commit": "9dac26a", + "tps": 281.88, + "top1": 0.9553, + "kl": 0.028, + "frontier_tps": 273.0, + "label": "XS", + "delta_tps": 8.88, + "pct_over_frontier": 3.3, + "pct_of_llama": 3.2, + "pct_of_ceiling": 77.0, + "effective_pct": 3.2, + "pass": true, + "clocks_pinned": false, + "clock_mhz": "0", + "clock_spread_mhz": "0", + "pin_target_mhz": "", + "eval_seed": "74c3316f061a588d", + "model_sha_pinned": true, + "llama_commit": "6f4f53f2b7da54fcdbbecaaa734337c337ad6176", + "eval_mode": "longctx", + "decode_tokens": 128, + "score_context": 128, + "best_context_label": "128-context", + "context_gains_pct": { + "128-context": 3.25, + "512-context": 3.01, + "4k-context": 2.67 + }, + "regression_labels": [], + "guard_context": 0, + "ctx_128_tps": 281.88, + "ctx_512_tps": 278.27, + "ctx_4096_tps": 270.05, + "ctx_16384_tps": 0.0, + "ctx_32768_tps": 0.0, + "guard_128_baseline": 273.0, + "guard_128_ratio": 1.0325, + "guard_128_tol": 0.98, + "guard_128_pass": true, + "guard_512_baseline": 270.13, + "guard_512_ratio": 1.0301, + "guard_512_tol": 0.98, + "guard_512_pass": true, + "guard_4k_baseline": 263.02, + "guard_4k_ratio": 1.0267, + "guard_4k_tol": 0.98, + "guard_4k_pass": true, + "guard_16k_baseline": 0.0, + "guard_16k_ratio": 0.0, + "guard_16k_tol": 0.98, + "guard_16k_pass": true, + "guard_32k_baseline": 0.0, + "guard_32k_ratio": 0.0, + "guard_32k_tol": 0.98, + "guard_32k_pass": true, + "model": "Qwythos-9B (Q4_K_M)", + "guard_model": "Qwen3.6-35B-A3B", + "guard": { + "pass": true, + "top1": 0.9665, + "kl": 0.0203, + "label": "M", + "ctx_128_tps": 427.78, + "ctx_512_tps": 421.19, + "ctx_4096_tps": 404.51, + "ctx_16384_tps": 387.65, + "ctx_32768_tps": 365.4, + "guard_128_pass": true, + "guard_512_pass": true, + "guard_4k_pass": true, + "guard_16k_pass": true, + "guard_32k_pass": true, + "speed_ok": true, + "accuracy_ok": true + } + }, + "score_qwen36": { + "commit": "9dac26a", + "tps": 365.46, + "top1": 0.9665, + "kl": 0.0203, + "frontier_tps": 353.02, + "label": "S", + "delta_tps": 12.44, + "pct_over_frontier": 3.5, + "pct_of_llama": 3.5, + "pct_of_ceiling": 99.9, + "effective_pct": 3.5, + "pass": true, + "clocks_pinned": false, + "clock_mhz": "0", + "clock_spread_mhz": "0", + "pin_target_mhz": "", + "eval_seed": "74c3316f061a588d", + "model_sha_pinned": true, + "llama_commit": "6f4f53f2b7da54fcdbbecaaa734337c337ad6176", + "eval_mode": "longctx", + "decode_tokens": 128, + "score_context": 32768, + "best_context_label": "32k-context", + "context_gains_pct": { + "128-context": 0.32, + "512-context": -0.1, + "4k-context": -0.11, + "16k-context": 1.81, + "32k-context": 3.52 + }, + "regression_labels": [], + "guard_context": 0, + "ctx_128_tps": 427.92, + "ctx_512_tps": 421.15, + "ctx_4096_tps": 404.53, + "ctx_16384_tps": 387.58, + "ctx_32768_tps": 365.46, + "guard_128_baseline": 426.56, + "guard_128_ratio": 1.0032, + "guard_128_tol": 0.98, + "guard_128_pass": true, + "guard_512_baseline": 421.57, + "guard_512_ratio": 0.999, + "guard_512_tol": 0.98, + "guard_512_pass": true, + "guard_4k_baseline": 404.98, + "guard_4k_ratio": 0.9989, + "guard_4k_tol": 0.98, + "guard_4k_pass": true, + "guard_16k_baseline": 380.69, + "guard_16k_ratio": 1.0181, + "guard_16k_tol": 0.98, + "guard_16k_pass": true, + "guard_32k_baseline": 353.02, + "guard_32k_ratio": 1.0352, + "guard_32k_tol": 0.98, + "guard_32k_pass": true, + "model": "Qwen3.6-35B-A3B", + "guard_model": "Qwythos-9B (Q4_K_M)", + "guard": { + "pass": true, + "top1": 0.9553, + "kl": 0.028, + "label": "XS", + "ctx_128_tps": 281.54, + "ctx_512_tps": 277.84, + "ctx_4096_tps": 269.82, + "ctx_16384_tps": 0.0, + "ctx_32768_tps": 0.0, + "guard_128_pass": true, + "guard_512_pass": true, + "guard_4k_pass": true, + "guard_16k_pass": true, + "guard_32k_pass": true, + "speed_ok": true, + "accuracy_ok": true + } + }, + "ctx_128_tps": 427.92, + "ctx_512_tps": 421.15, + "ctx_4096_tps": 404.53, + "ctx_16384_tps": 387.58, + "ctx_32768_tps": 365.46, + "guard_128_baseline": 426.56, + "guard_128_ratio": 1.0032, + "guard_128_pass": true, + "guard_512_baseline": 421.57, + "guard_512_ratio": 0.999, + "guard_512_pass": true, + "guard_4k_baseline": 404.98, + "guard_4k_ratio": 0.9989, + "guard_4k_pass": true, + "guard_16k_baseline": 380.69, + "guard_16k_ratio": 1.0181, + "guard_16k_pass": true, + "guard_32k_baseline": 353.02, + "guard_32k_ratio": 1.0352, + "guard_32k_pass": true, + "proof_url": "https://gittensor-ai-lab.github.io/sparkinfer-log/?run=0331-9dac26a", + "proof_run": "0331-9dac26a" + }, + { + "num": 294, + "title": "perf: Qwen3.6 +36 tok/s @12k, Qwythos 264 tok/s (beats llama.cpp 225)", + "areas": [ + "kernels", + "runtime" + ], + "label": "REJECT", + "tps": 370.91, + "delta_pct": 5.1, + "top1": 0.9713, + "kl": 0.01, + "url": "https://github.com/gittensor-ai-lab/sparkinfer/pull/294", + "model": "bidir", + "polaris_receipt_url": "https://gittensor-ai-lab.github.io/sparkinfer-log/?run=0294-f9065a6", + "polaris_receipt_hash": "3dfc54cb7104c57a", + "eval_mode": "longctx", + "score_context": 32768, + "best_context_label": "32k-context", + "context_gains_pct": { + "128-context": 0.08, + "512-context": 0.15, + "4k-context": 0.13, + "16k-context": 3.39, + "32k-context": 5.07 + }, + "regression_labels": [ + "regression-qwen35-128", + "regression-qwen35-512", + "regression-qwen35-4k" + ], + "mode": "bidir", + "label_qwen35": "REJECT", + "label_qwen36": "REJECT", + "pass_qwen35": false, + "pass_qwen36": false, + "score_qwen35": { + "commit": "f9065a6", + "tps": 255.74, + "top1": 0.9655, + "kl": 0.0125, + "frontier_tps": 270.13, + "label": "REJECT", + "pass": false, + "reason": "128-token decode no-regression gate: 258.37 tok/s < 98% of main 273.00 tok/s; 512-context decode no-regression gate: 255.74 tok/s < 98% of main 270.13 tok/s; 4k-context decode no-regression gate: 243.63 tok/s < 98% of main 263.02 tok/s", + "auto_close": true, + "delta_tps": -14.39, + "pct_over_frontier": -5.3, + "clocks_pinned": false, + "clock_mhz": "0", + "clock_spread_mhz": "0", + "pin_target_mhz": "", + "eval_seed": "b98b61fe6737c818", + "model_sha_pinned": true, + "llama_commit": "6f4f53f2b7da54fcdbbecaaa734337c337ad6176", + "eval_mode": "longctx", + "decode_tokens": 128, + "score_context": 512, + "best_context_label": "512-context", + "context_gains_pct": { + "128-context": -5.36, + "512-context": -5.33, + "4k-context": -7.37 + }, + "regression_labels": [ + "regression-128", + "regression-512", + "regression-4k" + ], + "guard_context": 0, + "ctx_128_tps": 258.37, + "ctx_512_tps": 255.74, + "ctx_4096_tps": 243.63, + "ctx_16384_tps": 0.0, + "ctx_32768_tps": 0.0, + "guard_128_baseline": 273.0, + "guard_128_ratio": 0.9464, + "guard_128_tol": 0.98, + "guard_128_pass": false, + "guard_512_baseline": 270.13, + "guard_512_ratio": 0.9467, + "guard_512_tol": 0.98, + "guard_512_pass": false, + "guard_4k_baseline": 263.02, + "guard_4k_ratio": 0.9263, + "guard_4k_tol": 0.98, + "guard_4k_pass": false, + "guard_16k_baseline": 0.0, + "guard_16k_ratio": 0.0, + "guard_16k_tol": 0.98, + "guard_16k_pass": true, + "guard_32k_baseline": 0.0, + "guard_32k_ratio": 0.0, + "guard_32k_tol": 0.98, + "guard_32k_pass": true, + "model": "Qwythos-9B (Q4_K_M)", + "guard_model": "Qwen3.6-35B-A3B", + "guard": { + "pass": true, + "top1": 0.9713, + "kl": 0.01, + "label": "M", + "ctx_128_tps": 426.77, + "ctx_512_tps": 422.13, + "ctx_4096_tps": 405.46, + "ctx_16384_tps": 393.45, + "ctx_32768_tps": 370.45, + "guard_128_pass": true, + "guard_512_pass": true, + "guard_4k_pass": true, + "guard_16k_pass": true, + "guard_32k_pass": true, + "speed_ok": true, + "accuracy_ok": true + } + }, + "score_qwen36": { + "commit": "f9065a6", + "tps": 370.91, + "top1": 0.9713, + "kl": 0.01, + "frontier_tps": 353.02, + "label": "REJECT", + "delta_tps": 17.89, + "pct_over_frontier": 5.1, + "pct_of_llama": 5.1, + "pct_of_ceiling": 101.3, + "effective_pct": 5.1, + "pass": false, + "clocks_pinned": false, + "clock_mhz": "0", + "clock_spread_mhz": "0", + "pin_target_mhz": "", + "eval_seed": "b98b61fe6737c818", + "model_sha_pinned": true, + "llama_commit": "6f4f53f2b7da54fcdbbecaaa734337c337ad6176", + "eval_mode": "longctx", + "decode_tokens": 128, + "score_context": 32768, + "best_context_label": "32k-context", + "context_gains_pct": { + "128-context": 0.08, + "512-context": 0.15, + "4k-context": 0.13, + "16k-context": 3.39, + "32k-context": 5.07 + }, + "regression_labels": [], + "guard_context": 0, + "ctx_128_tps": 426.88, + "ctx_512_tps": 422.2, + "ctx_4096_tps": 405.51, + "ctx_16384_tps": 393.58, + "ctx_32768_tps": 370.91, + "guard_128_baseline": 426.56, + "guard_128_ratio": 1.0008, + "guard_128_tol": 0.98, + "guard_128_pass": true, + "guard_512_baseline": 421.57, + "guard_512_ratio": 1.0015, + "guard_512_tol": 0.98, + "guard_512_pass": true, + "guard_4k_baseline": 404.98, + "guard_4k_ratio": 1.0013, + "guard_4k_tol": 0.98, + "guard_4k_pass": true, + "guard_16k_baseline": 380.69, + "guard_16k_ratio": 1.0339, + "guard_16k_tol": 0.98, + "guard_16k_pass": true, + "guard_32k_baseline": 353.02, + "guard_32k_ratio": 1.0507, + "guard_32k_tol": 0.98, + "guard_32k_pass": true, + "model": "Qwen3.6-35B-A3B", + "guard_model": "Qwythos-9B (Q4_K_M)", + "guard": { + "pass": false, + "top1": 0.9655, + "kl": 0.0125, + "label": "REJECT", + "ctx_128_tps": 258.29, + "ctx_512_tps": 255.69, + "ctx_4096_tps": 243.67, + "ctx_16384_tps": 0.0, + "ctx_32768_tps": 0.0, + "guard_128_pass": false, + "guard_512_pass": false, + "guard_4k_pass": false, + "guard_16k_pass": true, + "guard_32k_pass": true, + "speed_ok": false, + "accuracy_ok": true + }, + "reason": "no-regression guard: Qwythos-9B (Q4_K_M) decode regressed at: 128, 512, 4k", + "guard_regression_labels": [ + "regression-qwen35-128", + "regression-qwen35-512", + "regression-qwen35-4k" + ] + }, + "ctx_128_tps": 426.88, + "ctx_512_tps": 422.2, + "ctx_4096_tps": 405.51, + "ctx_16384_tps": 393.58, + "ctx_32768_tps": 370.91, + "guard_128_baseline": 426.56, + "guard_128_ratio": 1.0008, + "guard_128_pass": true, + "guard_512_baseline": 421.57, + "guard_512_ratio": 1.0015, + "guard_512_pass": true, + "guard_4k_baseline": 404.98, + "guard_4k_ratio": 1.0013, + "guard_4k_pass": true, + "guard_16k_baseline": 380.69, + "guard_16k_ratio": 1.0339, + "guard_16k_pass": true, + "guard_32k_baseline": 353.02, + "guard_32k_ratio": 1.0507, + "guard_32k_pass": true, + "proof_url": "https://gittensor-ai-lab.github.io/sparkinfer-log/?run=0294-f9065a6", + "proof_run": "0294-f9065a6" + }, { "num": 326, "title": "perf(attn): GQA-4 shared-KV tile for Qwythos hd256 full-attn decode (+2.4% @4k)", @@ -972,213 +1404,6 @@ window.SPARKINFER = { "proof_url": "https://gittensor-ai-lab.github.io/sparkinfer-log/?run=0318-99a48b6", "proof_run": "0318-99a48b6" }, - { - "num": 294, - "title": "perf: Qwen3.6 +36 tok/s @12k, Qwythos 264 tok/s (beats llama.cpp 225)", - "areas": [ - "kernels", - "runtime" - ], - "label": "none", - "tps": 393.21, - "delta_pct": 1.9, - "top1": 0.9752, - "kl": 0.0171, - "url": "https://github.com/gittensor-ai-lab/sparkinfer/pull/294", - "model": "bidir", - "polaris_receipt_url": "https://gittensor-ai-lab.github.io/sparkinfer-log/?run=0294-2aca228", - "polaris_receipt_hash": "0fb16ab5f47ecff5", - "eval_mode": "longctx", - "score_context": 16384, - "best_context_label": "16k-context", - "context_gains_pct": { - "128-context": 0.1, - "512-context": 0.55, - "4k-context": 0.51, - "16k-context": 1.94, - "32k-context": 1.84 - }, - "regression_labels": [], - "mode": "bidir", - "label_qwen35": "none", - "label_qwen36": "none", - "pass_qwen35": true, - "pass_qwen36": true, - "score_qwen35": { - "commit": "2aca228", - "tps": 255.63, - "top1": 0.9876, - "kl": 0.0159, - "frontier_tps": 254.27, - "label": "none", - "delta_tps": 1.36, - "pct_over_frontier": 0.5, - "note": "within significance gate — not a verified improvement", - "pass": true, - "clocks_pinned": false, - "clock_mhz": "0", - "clock_spread_mhz": "0", - "pin_target_mhz": "", - "eval_seed": "eab5246a3f500a9a", - "model_sha_pinned": true, - "llama_commit": "6f4f53f2b7da54fcdbbecaaa734337c337ad6176", - "eval_mode": "longctx", - "decode_tokens": 128, - "score_context": 512, - "best_context_label": "512-context", - "context_gains_pct": { - "128-context": 0.31, - "512-context": 0.53, - "4k-context": 0.42 - }, - "regression_labels": [], - "guard_context": 0, - "ctx_128_tps": 258.27, - "ctx_512_tps": 255.63, - "ctx_4096_tps": 243.52, - "ctx_16384_tps": 0.0, - "ctx_32768_tps": 0.0, - "guard_128_baseline": 257.47, - "guard_128_ratio": 1.0031, - "guard_128_tol": 0.98, - "guard_128_pass": true, - "guard_512_baseline": 254.27, - "guard_512_ratio": 1.0053, - "guard_512_tol": 0.98, - "guard_512_pass": true, - "guard_4k_baseline": 242.49, - "guard_4k_ratio": 1.0042, - "guard_4k_tol": 0.98, - "guard_4k_pass": true, - "guard_16k_baseline": 0.0, - "guard_16k_ratio": 0.0, - "guard_16k_tol": 0.98, - "guard_16k_pass": true, - "guard_32k_baseline": 0.0, - "guard_32k_ratio": 0.0, - "guard_32k_tol": 0.98, - "guard_32k_pass": true, - "model": "Qwythos-9B (Q4_K_M)", - "guard_model": "Qwen3.6-35B-A3B", - "guard": { - "pass": true, - "top1": 0.9752, - "kl": 0.0171, - "label": "none", - "ctx_128_tps": 426.77, - "ctx_512_tps": 421.97, - "ctx_4096_tps": 405.12, - "ctx_16384_tps": 393.14, - "ctx_32768_tps": 370.31, - "guard_128_pass": true, - "guard_512_pass": true, - "guard_4k_pass": true, - "guard_16k_pass": true, - "guard_32k_pass": true, - "speed_ok": true, - "accuracy_ok": true - } - }, - "score_qwen36": { - "commit": "2aca228", - "tps": 393.21, - "top1": 0.9752, - "kl": 0.0171, - "frontier_tps": 385.74, - "label": "none", - "delta_tps": 7.47, - "pct_over_frontier": 1.9, - "note": "within significance gate — not a verified improvement", - "pass": true, - "clocks_pinned": false, - "clock_mhz": "0", - "clock_spread_mhz": "0", - "pin_target_mhz": "", - "eval_seed": "eab5246a3f500a9a", - "model_sha_pinned": true, - "llama_commit": "6f4f53f2b7da54fcdbbecaaa734337c337ad6176", - "eval_mode": "longctx", - "decode_tokens": 128, - "score_context": 16384, - "best_context_label": "16k-context", - "context_gains_pct": { - "128-context": 0.1, - "512-context": 0.55, - "4k-context": 0.51, - "16k-context": 1.94, - "32k-context": 1.84 - }, - "regression_labels": [], - "guard_context": 0, - "ctx_128_tps": 426.89, - "ctx_512_tps": 422.23, - "ctx_4096_tps": 405.08, - "ctx_16384_tps": 393.21, - "ctx_32768_tps": 370.6, - "guard_128_baseline": 426.48, - "guard_128_ratio": 1.001, - "guard_128_tol": 0.98, - "guard_128_pass": true, - "guard_512_baseline": 419.93, - "guard_512_ratio": 1.0055, - "guard_512_tol": 0.98, - "guard_512_pass": true, - "guard_4k_baseline": 403.01, - "guard_4k_ratio": 1.0051, - "guard_4k_tol": 0.98, - "guard_4k_pass": true, - "guard_16k_baseline": 385.74, - "guard_16k_ratio": 1.0194, - "guard_16k_tol": 0.98, - "guard_16k_pass": true, - "guard_32k_baseline": 363.89, - "guard_32k_ratio": 1.0184, - "guard_32k_tol": 0.98, - "guard_32k_pass": true, - "model": "Qwen3.6-35B-A3B", - "guard_model": "Qwythos-9B (Q4_K_M)", - "guard": { - "pass": true, - "top1": 0.9876, - "kl": 0.0159, - "label": "none", - "ctx_128_tps": 258.05, - "ctx_512_tps": 255.4, - "ctx_4096_tps": 243.37, - "ctx_16384_tps": 0.0, - "ctx_32768_tps": 0.0, - "guard_128_pass": true, - "guard_512_pass": true, - "guard_4k_pass": true, - "guard_16k_pass": true, - "guard_32k_pass": true, - "speed_ok": true, - "accuracy_ok": true - } - }, - "ctx_128_tps": 426.89, - "ctx_512_tps": 422.23, - "ctx_4096_tps": 405.08, - "ctx_16384_tps": 393.21, - "ctx_32768_tps": 370.6, - "guard_128_baseline": 426.48, - "guard_128_ratio": 1.001, - "guard_128_pass": true, - "guard_512_baseline": 419.93, - "guard_512_ratio": 1.0055, - "guard_512_pass": true, - "guard_4k_baseline": 403.01, - "guard_4k_ratio": 1.0051, - "guard_4k_pass": true, - "guard_16k_baseline": 385.74, - "guard_16k_ratio": 1.0194, - "guard_16k_pass": true, - "guard_32k_baseline": 363.89, - "guard_32k_ratio": 1.0184, - "guard_32k_pass": true, - "proof_url": "https://gittensor-ai-lab.github.io/sparkinfer-log/?run=0294-2aca228", - "proof_run": "0294-2aca228" - }, { "num": 323, "title": "perf(qwen35): requantize dense FFN down Q6_K→Q4_K at load (~5% Qwythos decode)", @@ -2589,35 +2814,6 @@ window.SPARKINFER = { "kl": 13.9592, "url": "https://github.com/gittensor-ai-lab/sparkinfer/pull/117", "proof_url": "https://gittensor-ai-lab.github.io/sparkinfer-log/?run=0117-2c0848b" - }, - { - "num": 91, - "title": "feat(moe): LRU expert cache + async layer-ahead prefetch", - "areas": [ - "moe", - "runtime" - ], - "label": "REJECT", - "tps": 0, - "delta_pct": null, - "top1": 0, - "kl": 99, - "url": "https://github.com/gittensor-ai-lab/sparkinfer/pull/91", - "proof_url": "https://gittensor-ai-lab.github.io/sparkinfer-log/?run=0091-cf9803d" - }, - { - "num": 114, - "title": "perf(runtime): context-adaptive flash-decode split count", - "areas": [ - "runtime" - ], - "label": "none", - "tps": 462.02, - "delta_pct": 0.1, - "top1": 0.9851, - "kl": 0.0093, - "url": "https://github.com/gittensor-ai-lab/sparkinfer/pull/114", - "proof_url": "https://gittensor-ai-lab.github.io/sparkinfer-log/?run=0114-a534177" } ], "landed": [ diff --git a/dashboard/data.json b/dashboard/data.json index 572365b..ac1b2e8 100644 --- a/dashboard/data.json +++ b/dashboard/data.json @@ -136,6 +136,438 @@ } ], "prs": [ + { + "num": 331, + "title": "perf(moe): complete gate_up->quant_h->down PDL chain for bs=1 decode", + "areas": [ + "kernels" + ], + "label": "S", + "tps": 365.46, + "delta_pct": 3.5, + "top1": 0.9665, + "kl": 0.0203, + "url": "https://github.com/gittensor-ai-lab/sparkinfer/pull/331", + "model": "bidir", + "eval_mode": "longctx", + "score_context": 32768, + "best_context_label": "32k-context", + "context_gains_pct": { + "128-context": 0.32, + "512-context": -0.1, + "4k-context": -0.11, + "16k-context": 1.81, + "32k-context": 3.52 + }, + "regression_labels": [], + "mode": "bidir", + "label_qwen35": "XS", + "label_qwen36": "S", + "pass_qwen35": true, + "pass_qwen36": true, + "score_qwen35": { + "commit": "9dac26a", + "tps": 281.88, + "top1": 0.9553, + "kl": 0.028, + "frontier_tps": 273.0, + "label": "XS", + "delta_tps": 8.88, + "pct_over_frontier": 3.3, + "pct_of_llama": 3.2, + "pct_of_ceiling": 77.0, + "effective_pct": 3.2, + "pass": true, + "clocks_pinned": false, + "clock_mhz": "0", + "clock_spread_mhz": "0", + "pin_target_mhz": "", + "eval_seed": "74c3316f061a588d", + "model_sha_pinned": true, + "llama_commit": "6f4f53f2b7da54fcdbbecaaa734337c337ad6176", + "eval_mode": "longctx", + "decode_tokens": 128, + "score_context": 128, + "best_context_label": "128-context", + "context_gains_pct": { + "128-context": 3.25, + "512-context": 3.01, + "4k-context": 2.67 + }, + "regression_labels": [], + "guard_context": 0, + "ctx_128_tps": 281.88, + "ctx_512_tps": 278.27, + "ctx_4096_tps": 270.05, + "ctx_16384_tps": 0.0, + "ctx_32768_tps": 0.0, + "guard_128_baseline": 273.0, + "guard_128_ratio": 1.0325, + "guard_128_tol": 0.98, + "guard_128_pass": true, + "guard_512_baseline": 270.13, + "guard_512_ratio": 1.0301, + "guard_512_tol": 0.98, + "guard_512_pass": true, + "guard_4k_baseline": 263.02, + "guard_4k_ratio": 1.0267, + "guard_4k_tol": 0.98, + "guard_4k_pass": true, + "guard_16k_baseline": 0.0, + "guard_16k_ratio": 0.0, + "guard_16k_tol": 0.98, + "guard_16k_pass": true, + "guard_32k_baseline": 0.0, + "guard_32k_ratio": 0.0, + "guard_32k_tol": 0.98, + "guard_32k_pass": true, + "model": "Qwythos-9B (Q4_K_M)", + "guard_model": "Qwen3.6-35B-A3B", + "guard": { + "pass": true, + "top1": 0.9665, + "kl": 0.0203, + "label": "M", + "ctx_128_tps": 427.78, + "ctx_512_tps": 421.19, + "ctx_4096_tps": 404.51, + "ctx_16384_tps": 387.65, + "ctx_32768_tps": 365.4, + "guard_128_pass": true, + "guard_512_pass": true, + "guard_4k_pass": true, + "guard_16k_pass": true, + "guard_32k_pass": true, + "speed_ok": true, + "accuracy_ok": true + } + }, + "score_qwen36": { + "commit": "9dac26a", + "tps": 365.46, + "top1": 0.9665, + "kl": 0.0203, + "frontier_tps": 353.02, + "label": "S", + "delta_tps": 12.44, + "pct_over_frontier": 3.5, + "pct_of_llama": 3.5, + "pct_of_ceiling": 99.9, + "effective_pct": 3.5, + "pass": true, + "clocks_pinned": false, + "clock_mhz": "0", + "clock_spread_mhz": "0", + "pin_target_mhz": "", + "eval_seed": "74c3316f061a588d", + "model_sha_pinned": true, + "llama_commit": "6f4f53f2b7da54fcdbbecaaa734337c337ad6176", + "eval_mode": "longctx", + "decode_tokens": 128, + "score_context": 32768, + "best_context_label": "32k-context", + "context_gains_pct": { + "128-context": 0.32, + "512-context": -0.1, + "4k-context": -0.11, + "16k-context": 1.81, + "32k-context": 3.52 + }, + "regression_labels": [], + "guard_context": 0, + "ctx_128_tps": 427.92, + "ctx_512_tps": 421.15, + "ctx_4096_tps": 404.53, + "ctx_16384_tps": 387.58, + "ctx_32768_tps": 365.46, + "guard_128_baseline": 426.56, + "guard_128_ratio": 1.0032, + "guard_128_tol": 0.98, + "guard_128_pass": true, + "guard_512_baseline": 421.57, + "guard_512_ratio": 0.999, + "guard_512_tol": 0.98, + "guard_512_pass": true, + "guard_4k_baseline": 404.98, + "guard_4k_ratio": 0.9989, + "guard_4k_tol": 0.98, + "guard_4k_pass": true, + "guard_16k_baseline": 380.69, + "guard_16k_ratio": 1.0181, + "guard_16k_tol": 0.98, + "guard_16k_pass": true, + "guard_32k_baseline": 353.02, + "guard_32k_ratio": 1.0352, + "guard_32k_tol": 0.98, + "guard_32k_pass": true, + "model": "Qwen3.6-35B-A3B", + "guard_model": "Qwythos-9B (Q4_K_M)", + "guard": { + "pass": true, + "top1": 0.9553, + "kl": 0.028, + "label": "XS", + "ctx_128_tps": 281.54, + "ctx_512_tps": 277.84, + "ctx_4096_tps": 269.82, + "ctx_16384_tps": 0.0, + "ctx_32768_tps": 0.0, + "guard_128_pass": true, + "guard_512_pass": true, + "guard_4k_pass": true, + "guard_16k_pass": true, + "guard_32k_pass": true, + "speed_ok": true, + "accuracy_ok": true + } + }, + "ctx_128_tps": 427.92, + "ctx_512_tps": 421.15, + "ctx_4096_tps": 404.53, + "ctx_16384_tps": 387.58, + "ctx_32768_tps": 365.46, + "guard_128_baseline": 426.56, + "guard_128_ratio": 1.0032, + "guard_128_pass": true, + "guard_512_baseline": 421.57, + "guard_512_ratio": 0.999, + "guard_512_pass": true, + "guard_4k_baseline": 404.98, + "guard_4k_ratio": 0.9989, + "guard_4k_pass": true, + "guard_16k_baseline": 380.69, + "guard_16k_ratio": 1.0181, + "guard_16k_pass": true, + "guard_32k_baseline": 353.02, + "guard_32k_ratio": 1.0352, + "guard_32k_pass": true, + "proof_url": "https://gittensor-ai-lab.github.io/sparkinfer-log/?run=0331-9dac26a", + "proof_run": "0331-9dac26a" + }, + { + "num": 294, + "title": "perf: Qwen3.6 +36 tok/s @12k, Qwythos 264 tok/s (beats llama.cpp 225)", + "areas": [ + "kernels", + "runtime" + ], + "label": "REJECT", + "tps": 370.91, + "delta_pct": 5.1, + "top1": 0.9713, + "kl": 0.01, + "url": "https://github.com/gittensor-ai-lab/sparkinfer/pull/294", + "model": "bidir", + "polaris_receipt_url": "https://gittensor-ai-lab.github.io/sparkinfer-log/?run=0294-f9065a6", + "polaris_receipt_hash": "3dfc54cb7104c57a", + "eval_mode": "longctx", + "score_context": 32768, + "best_context_label": "32k-context", + "context_gains_pct": { + "128-context": 0.08, + "512-context": 0.15, + "4k-context": 0.13, + "16k-context": 3.39, + "32k-context": 5.07 + }, + "regression_labels": [ + "regression-qwen35-128", + "regression-qwen35-512", + "regression-qwen35-4k" + ], + "mode": "bidir", + "label_qwen35": "REJECT", + "label_qwen36": "REJECT", + "pass_qwen35": false, + "pass_qwen36": false, + "score_qwen35": { + "commit": "f9065a6", + "tps": 255.74, + "top1": 0.9655, + "kl": 0.0125, + "frontier_tps": 270.13, + "label": "REJECT", + "pass": false, + "reason": "128-token decode no-regression gate: 258.37 tok/s < 98% of main 273.00 tok/s; 512-context decode no-regression gate: 255.74 tok/s < 98% of main 270.13 tok/s; 4k-context decode no-regression gate: 243.63 tok/s < 98% of main 263.02 tok/s", + "auto_close": true, + "delta_tps": -14.39, + "pct_over_frontier": -5.3, + "clocks_pinned": false, + "clock_mhz": "0", + "clock_spread_mhz": "0", + "pin_target_mhz": "", + "eval_seed": "b98b61fe6737c818", + "model_sha_pinned": true, + "llama_commit": "6f4f53f2b7da54fcdbbecaaa734337c337ad6176", + "eval_mode": "longctx", + "decode_tokens": 128, + "score_context": 512, + "best_context_label": "512-context", + "context_gains_pct": { + "128-context": -5.36, + "512-context": -5.33, + "4k-context": -7.37 + }, + "regression_labels": [ + "regression-128", + "regression-512", + "regression-4k" + ], + "guard_context": 0, + "ctx_128_tps": 258.37, + "ctx_512_tps": 255.74, + "ctx_4096_tps": 243.63, + "ctx_16384_tps": 0.0, + "ctx_32768_tps": 0.0, + "guard_128_baseline": 273.0, + "guard_128_ratio": 0.9464, + "guard_128_tol": 0.98, + "guard_128_pass": false, + "guard_512_baseline": 270.13, + "guard_512_ratio": 0.9467, + "guard_512_tol": 0.98, + "guard_512_pass": false, + "guard_4k_baseline": 263.02, + "guard_4k_ratio": 0.9263, + "guard_4k_tol": 0.98, + "guard_4k_pass": false, + "guard_16k_baseline": 0.0, + "guard_16k_ratio": 0.0, + "guard_16k_tol": 0.98, + "guard_16k_pass": true, + "guard_32k_baseline": 0.0, + "guard_32k_ratio": 0.0, + "guard_32k_tol": 0.98, + "guard_32k_pass": true, + "model": "Qwythos-9B (Q4_K_M)", + "guard_model": "Qwen3.6-35B-A3B", + "guard": { + "pass": true, + "top1": 0.9713, + "kl": 0.01, + "label": "M", + "ctx_128_tps": 426.77, + "ctx_512_tps": 422.13, + "ctx_4096_tps": 405.46, + "ctx_16384_tps": 393.45, + "ctx_32768_tps": 370.45, + "guard_128_pass": true, + "guard_512_pass": true, + "guard_4k_pass": true, + "guard_16k_pass": true, + "guard_32k_pass": true, + "speed_ok": true, + "accuracy_ok": true + } + }, + "score_qwen36": { + "commit": "f9065a6", + "tps": 370.91, + "top1": 0.9713, + "kl": 0.01, + "frontier_tps": 353.02, + "label": "REJECT", + "delta_tps": 17.89, + "pct_over_frontier": 5.1, + "pct_of_llama": 5.1, + "pct_of_ceiling": 101.3, + "effective_pct": 5.1, + "pass": false, + "clocks_pinned": false, + "clock_mhz": "0", + "clock_spread_mhz": "0", + "pin_target_mhz": "", + "eval_seed": "b98b61fe6737c818", + "model_sha_pinned": true, + "llama_commit": "6f4f53f2b7da54fcdbbecaaa734337c337ad6176", + "eval_mode": "longctx", + "decode_tokens": 128, + "score_context": 32768, + "best_context_label": "32k-context", + "context_gains_pct": { + "128-context": 0.08, + "512-context": 0.15, + "4k-context": 0.13, + "16k-context": 3.39, + "32k-context": 5.07 + }, + "regression_labels": [], + "guard_context": 0, + "ctx_128_tps": 426.88, + "ctx_512_tps": 422.2, + "ctx_4096_tps": 405.51, + "ctx_16384_tps": 393.58, + "ctx_32768_tps": 370.91, + "guard_128_baseline": 426.56, + "guard_128_ratio": 1.0008, + "guard_128_tol": 0.98, + "guard_128_pass": true, + "guard_512_baseline": 421.57, + "guard_512_ratio": 1.0015, + "guard_512_tol": 0.98, + "guard_512_pass": true, + "guard_4k_baseline": 404.98, + "guard_4k_ratio": 1.0013, + "guard_4k_tol": 0.98, + "guard_4k_pass": true, + "guard_16k_baseline": 380.69, + "guard_16k_ratio": 1.0339, + "guard_16k_tol": 0.98, + "guard_16k_pass": true, + "guard_32k_baseline": 353.02, + "guard_32k_ratio": 1.0507, + "guard_32k_tol": 0.98, + "guard_32k_pass": true, + "model": "Qwen3.6-35B-A3B", + "guard_model": "Qwythos-9B (Q4_K_M)", + "guard": { + "pass": false, + "top1": 0.9655, + "kl": 0.0125, + "label": "REJECT", + "ctx_128_tps": 258.29, + "ctx_512_tps": 255.69, + "ctx_4096_tps": 243.67, + "ctx_16384_tps": 0.0, + "ctx_32768_tps": 0.0, + "guard_128_pass": false, + "guard_512_pass": false, + "guard_4k_pass": false, + "guard_16k_pass": true, + "guard_32k_pass": true, + "speed_ok": false, + "accuracy_ok": true + }, + "reason": "no-regression guard: Qwythos-9B (Q4_K_M) decode regressed at: 128, 512, 4k", + "guard_regression_labels": [ + "regression-qwen35-128", + "regression-qwen35-512", + "regression-qwen35-4k" + ] + }, + "ctx_128_tps": 426.88, + "ctx_512_tps": 422.2, + "ctx_4096_tps": 405.51, + "ctx_16384_tps": 393.58, + "ctx_32768_tps": 370.91, + "guard_128_baseline": 426.56, + "guard_128_ratio": 1.0008, + "guard_128_pass": true, + "guard_512_baseline": 421.57, + "guard_512_ratio": 1.0015, + "guard_512_pass": true, + "guard_4k_baseline": 404.98, + "guard_4k_ratio": 1.0013, + "guard_4k_pass": true, + "guard_16k_baseline": 380.69, + "guard_16k_ratio": 1.0339, + "guard_16k_pass": true, + "guard_32k_baseline": 353.02, + "guard_32k_ratio": 1.0507, + "guard_32k_pass": true, + "proof_url": "https://gittensor-ai-lab.github.io/sparkinfer-log/?run=0294-f9065a6", + "proof_run": "0294-f9065a6" + }, { "num": 326, "title": "perf(attn): GQA-4 shared-KV tile for Qwythos hd256 full-attn decode (+2.4% @4k)", @@ -971,213 +1403,6 @@ "proof_url": "https://gittensor-ai-lab.github.io/sparkinfer-log/?run=0318-99a48b6", "proof_run": "0318-99a48b6" }, - { - "num": 294, - "title": "perf: Qwen3.6 +36 tok/s @12k, Qwythos 264 tok/s (beats llama.cpp 225)", - "areas": [ - "kernels", - "runtime" - ], - "label": "none", - "tps": 393.21, - "delta_pct": 1.9, - "top1": 0.9752, - "kl": 0.0171, - "url": "https://github.com/gittensor-ai-lab/sparkinfer/pull/294", - "model": "bidir", - "polaris_receipt_url": "https://gittensor-ai-lab.github.io/sparkinfer-log/?run=0294-2aca228", - "polaris_receipt_hash": "0fb16ab5f47ecff5", - "eval_mode": "longctx", - "score_context": 16384, - "best_context_label": "16k-context", - "context_gains_pct": { - "128-context": 0.1, - "512-context": 0.55, - "4k-context": 0.51, - "16k-context": 1.94, - "32k-context": 1.84 - }, - "regression_labels": [], - "mode": "bidir", - "label_qwen35": "none", - "label_qwen36": "none", - "pass_qwen35": true, - "pass_qwen36": true, - "score_qwen35": { - "commit": "2aca228", - "tps": 255.63, - "top1": 0.9876, - "kl": 0.0159, - "frontier_tps": 254.27, - "label": "none", - "delta_tps": 1.36, - "pct_over_frontier": 0.5, - "note": "within significance gate — not a verified improvement", - "pass": true, - "clocks_pinned": false, - "clock_mhz": "0", - "clock_spread_mhz": "0", - "pin_target_mhz": "", - "eval_seed": "eab5246a3f500a9a", - "model_sha_pinned": true, - "llama_commit": "6f4f53f2b7da54fcdbbecaaa734337c337ad6176", - "eval_mode": "longctx", - "decode_tokens": 128, - "score_context": 512, - "best_context_label": "512-context", - "context_gains_pct": { - "128-context": 0.31, - "512-context": 0.53, - "4k-context": 0.42 - }, - "regression_labels": [], - "guard_context": 0, - "ctx_128_tps": 258.27, - "ctx_512_tps": 255.63, - "ctx_4096_tps": 243.52, - "ctx_16384_tps": 0.0, - "ctx_32768_tps": 0.0, - "guard_128_baseline": 257.47, - "guard_128_ratio": 1.0031, - "guard_128_tol": 0.98, - "guard_128_pass": true, - "guard_512_baseline": 254.27, - "guard_512_ratio": 1.0053, - "guard_512_tol": 0.98, - "guard_512_pass": true, - "guard_4k_baseline": 242.49, - "guard_4k_ratio": 1.0042, - "guard_4k_tol": 0.98, - "guard_4k_pass": true, - "guard_16k_baseline": 0.0, - "guard_16k_ratio": 0.0, - "guard_16k_tol": 0.98, - "guard_16k_pass": true, - "guard_32k_baseline": 0.0, - "guard_32k_ratio": 0.0, - "guard_32k_tol": 0.98, - "guard_32k_pass": true, - "model": "Qwythos-9B (Q4_K_M)", - "guard_model": "Qwen3.6-35B-A3B", - "guard": { - "pass": true, - "top1": 0.9752, - "kl": 0.0171, - "label": "none", - "ctx_128_tps": 426.77, - "ctx_512_tps": 421.97, - "ctx_4096_tps": 405.12, - "ctx_16384_tps": 393.14, - "ctx_32768_tps": 370.31, - "guard_128_pass": true, - "guard_512_pass": true, - "guard_4k_pass": true, - "guard_16k_pass": true, - "guard_32k_pass": true, - "speed_ok": true, - "accuracy_ok": true - } - }, - "score_qwen36": { - "commit": "2aca228", - "tps": 393.21, - "top1": 0.9752, - "kl": 0.0171, - "frontier_tps": 385.74, - "label": "none", - "delta_tps": 7.47, - "pct_over_frontier": 1.9, - "note": "within significance gate — not a verified improvement", - "pass": true, - "clocks_pinned": false, - "clock_mhz": "0", - "clock_spread_mhz": "0", - "pin_target_mhz": "", - "eval_seed": "eab5246a3f500a9a", - "model_sha_pinned": true, - "llama_commit": "6f4f53f2b7da54fcdbbecaaa734337c337ad6176", - "eval_mode": "longctx", - "decode_tokens": 128, - "score_context": 16384, - "best_context_label": "16k-context", - "context_gains_pct": { - "128-context": 0.1, - "512-context": 0.55, - "4k-context": 0.51, - "16k-context": 1.94, - "32k-context": 1.84 - }, - "regression_labels": [], - "guard_context": 0, - "ctx_128_tps": 426.89, - "ctx_512_tps": 422.23, - "ctx_4096_tps": 405.08, - "ctx_16384_tps": 393.21, - "ctx_32768_tps": 370.6, - "guard_128_baseline": 426.48, - "guard_128_ratio": 1.001, - "guard_128_tol": 0.98, - "guard_128_pass": true, - "guard_512_baseline": 419.93, - "guard_512_ratio": 1.0055, - "guard_512_tol": 0.98, - "guard_512_pass": true, - "guard_4k_baseline": 403.01, - "guard_4k_ratio": 1.0051, - "guard_4k_tol": 0.98, - "guard_4k_pass": true, - "guard_16k_baseline": 385.74, - "guard_16k_ratio": 1.0194, - "guard_16k_tol": 0.98, - "guard_16k_pass": true, - "guard_32k_baseline": 363.89, - "guard_32k_ratio": 1.0184, - "guard_32k_tol": 0.98, - "guard_32k_pass": true, - "model": "Qwen3.6-35B-A3B", - "guard_model": "Qwythos-9B (Q4_K_M)", - "guard": { - "pass": true, - "top1": 0.9876, - "kl": 0.0159, - "label": "none", - "ctx_128_tps": 258.05, - "ctx_512_tps": 255.4, - "ctx_4096_tps": 243.37, - "ctx_16384_tps": 0.0, - "ctx_32768_tps": 0.0, - "guard_128_pass": true, - "guard_512_pass": true, - "guard_4k_pass": true, - "guard_16k_pass": true, - "guard_32k_pass": true, - "speed_ok": true, - "accuracy_ok": true - } - }, - "ctx_128_tps": 426.89, - "ctx_512_tps": 422.23, - "ctx_4096_tps": 405.08, - "ctx_16384_tps": 393.21, - "ctx_32768_tps": 370.6, - "guard_128_baseline": 426.48, - "guard_128_ratio": 1.001, - "guard_128_pass": true, - "guard_512_baseline": 419.93, - "guard_512_ratio": 1.0055, - "guard_512_pass": true, - "guard_4k_baseline": 403.01, - "guard_4k_ratio": 1.0051, - "guard_4k_pass": true, - "guard_16k_baseline": 385.74, - "guard_16k_ratio": 1.0194, - "guard_16k_pass": true, - "guard_32k_baseline": 363.89, - "guard_32k_ratio": 1.0184, - "guard_32k_pass": true, - "proof_url": "https://gittensor-ai-lab.github.io/sparkinfer-log/?run=0294-2aca228", - "proof_run": "0294-2aca228" - }, { "num": 323, "title": "perf(qwen35): requantize dense FFN down Q6_K→Q4_K at load (~5% Qwythos decode)", @@ -2588,35 +2813,6 @@ "kl": 13.9592, "url": "https://github.com/gittensor-ai-lab/sparkinfer/pull/117", "proof_url": "https://gittensor-ai-lab.github.io/sparkinfer-log/?run=0117-2c0848b" - }, - { - "num": 91, - "title": "feat(moe): LRU expert cache + async layer-ahead prefetch", - "areas": [ - "moe", - "runtime" - ], - "label": "REJECT", - "tps": 0, - "delta_pct": null, - "top1": 0, - "kl": 99, - "url": "https://github.com/gittensor-ai-lab/sparkinfer/pull/91", - "proof_url": "https://gittensor-ai-lab.github.io/sparkinfer-log/?run=0091-cf9803d" - }, - { - "num": 114, - "title": "perf(runtime): context-adaptive flash-decode split count", - "areas": [ - "runtime" - ], - "label": "none", - "tps": 462.02, - "delta_pct": 0.1, - "top1": 0.9851, - "kl": 0.0093, - "url": "https://github.com/gittensor-ai-lab/sparkinfer/pull/114", - "proof_url": "https://gittensor-ai-lab.github.io/sparkinfer-log/?run=0114-a534177" } ], "landed": [