Stage
Scored
12 of 12 stages
Settled in
11h 22m
committed Oct 15, 07:54 UTC
Round score
0.0912
0 is baseline speed
p99 latency
—
ceiling 2000 / 50 ms
Round 3: scored · 0.0912
12/12 stages
Intake & gates
6/6Patch hash recorded; awaiting fetch.
{
"hotkey": "5FKgasNZNKz62SgTmHuA9rahqiAnLQDoUMjUwy6okp9UEyNs",
"payment_tx": 33,
"commit_block": 9287353,
"payment_block": 9287350,
"patch_fingerprint": "sha256:221949b6eb7086bf8e04586cd334b6398fc4b4fe03e2b55ddc298454216e92b4"
}A worker claimed the submission and started the pipeline.
{
"job_id": 264
}Patch artifact retrieved from object storage.
{
"size": 298621,
"patch_hash": "sha256:6e14834691d0532c62d31c54adfeb7645dfd6939530ba9b38191b1452c81f8de"
}Integrity and path constraints checked.
{
"identity": "ok",
"patch_hash": "sha256:6e14834691d0532c62d31c54adfeb7645dfd6939530ba9b38191b1452c81f8de"
}Patch applied onto the baseline checkout.
{
"ok": true
}Public API / surface compatibility held.
{
"files": [
"python/sglang/kernels/jit/csrc/elementwise/set_mla_kv_concat_q.cuh",
"python/sglang/kernels/jit/csrc/kimi_k3/attn_res/fused_tma.cuh",
"python/sglang/kernels/jit/csrc/kimi_k3/comm/ar_fusion.cuh",
"python/sglang/kernels/jit/csrc/moe/route_quant_fused.cuh",
"python/sglang/kernels/jit/csrc/moe/route_radix.cuh",
"python/sglang/kernels/ops/attention/attn_res.py",
"python/sglang/kernels/ops/attention/dcp_kernels.py",
"python/sglang/kernels/ops/attention/kda_decode_mtp.py",
"python/sglang/kernels/ops/attention/set_mla_kv_concat_q.py",
"python/sglang/kernels/ops/communication/all_reduce_residual.py",
"python/sglang/kernels/ops/gemm/cutedsl_bf16_gemm.py",
"python/sglang/kernels/ops/mamba/mamba_state_scatter_triton.py",
"python/sglang/kernels/ops/moe/moe_route_quant_fused.py",
"python/sglang/kernels/ops/speculative/dspark/dspark_accept.py",
"python/sglang/srt/arg_groups/cuda_graph_hook.py",
"python/sglang/srt/arg_groups/fields/spec.py",
"python/sglang/srt/arg_groups/memory_hook.py",
"python/sglang/srt/arg_groups/model_overrides/kimi_k3.py",
"python/sglang/srt/arg_groups/speculative_hook.py",
"python/sglang/srt/entrypoints/openai/serving_chat.py",
"python/sglang/srt/environ.py",
"python/sglang/srt/layers/attention/linear/kda_backend.py",
"python/sglang/srt/layers/attention/trtllm_mla_backend.py",
"python/sglang/srt/layers/attn_residual.py",
"python/sglang/srt/layers/communication/k3_ar_fusion.py",
"python/sglang/srt/layers/dcp/comm.py",
"python/sglang/srt/layers/moe/moe_runner/flashinfer_trtllm.py",
"python/sglang/srt/layers/moe/route_quant_handoff.py",
"python/sglang/srt/layers/quantization/unquant.py",
"python/sglang/srt/managers/schedule_batch.py",
"python/sglang/srt/managers/scheduler.py",
"python/sglang/srt/managers/scheduler_components/internal_state.py",
"python/sglang/srt/managers/tokenizer_manager.py",
"python/sglang/srt/model_executor/forward_batch_info.py",
"python/sglang/srt/model_executor/runner/base_runner.py",
"python/sglang/srt/model_executor/runner/flashinfer_autotune.py",
"python/sglang/srt/models/deepseek_common/attention_forward_methods/forward_mla.py",
"python/sglang/srt/models/dspark.py",
"python/sglang/srt/models/kimi_k3.py",
"python/sglang/srt/speculative/dflash_worker_v2.py",
"python/sglang/srt/speculative/dspark_components/dspark_draft.py",
"python/sglang/srt/speculative/dspark_components/dspark_draft_sampler.py",
"python/sglang/srt/speculative/dspark_components/dspark_verify.py",
"python/sglang/srt/speculative/dspark_components/dspark_worker_v2.py",
"python/sglang/kernels/jit/csrc/elementwise/l2_prefetch.cuh",
"python/sglang/kernels/jit/csrc/moe/k3_decode_moe.cuh",
"python/sglang/kernels/ops/gemm/fp8_decode_shadow.py",
"python/sglang/kernels/ops/layernorm/dual_rmsnorm.py",
"python/sglang/kernels/ops/memory/l2_prefetch.py",
"python/sglang/kernels/ops/moe/k3_decode_moe.py",
"python/sglang/kernels/ops/speculative/dspark/commit_track_indices.py",
"python/sglang/kernels/ops/speculative/dspark/fused_accept_sampling.py",
"python/sglang/srt/layers/moe/moe_tactic_pins.py",
"python/sglang/srt/layers/moe/verify_route_alias.py",
"python/sglang/srt/speculative/dspark_components/dspark_verify_depth.py",
"python/sglang/srt/utils/prompt_ids_cache.py"
]
}Build
3/3Hermetic engine image build in progress.
No structured detail recorded for this step.
Candidate engine image pushed to the registry.
{
"image_ref": "ghcr.io/pareton-ai/pareton-engine@sha256:b528480d3838ab703b8e321fd6cc93055b014bf8cec7d2fae89706ca5ad133f9",
"image_digest": "sha256:b528480d3838ab703b8e321fd6cc93055b014bf8cec7d2fae89706ca5ad133f9"
}Candidate engine image built and digest pinned.
{
"mock": false,
"build_log": "/var/log/pareton/builds/feebc675-6958-44f2-8362-ae977f80657c/build.log",
"image_ref": "ghcr.io/pareton-ai/pareton-engine@sha256:b528480d3838ab703b8e321fd6cc93055b014bf8cec7d2fae89706ca5ad133f9",
"image_tag": "ghcr.io/pareton-ai/pareton-engine:6e14834691d0532c62d31c54adfeb7645dfd6939530ba9b38191b1452c81f8de"
}Round
3/3Waiting to be seated in a round.
No structured detail recorded for this step.
Seated in a round; waiting to run or running.
{
"ordinal": 3,
"round_id": "031ed7a0-e0e9-44b4-bce0-728750ab9eac"
}The image ran and holds a round score.
{
"score": 0.0911902325530142,
"reason": null,
"ordinal": 6,
"round_id": "4600ee4a-4b1e-404d-af1c-964ec63f3784",
"operator_recovery": {
"type": "operator_round_recovery",
"reason": "Operator accepts baseline repeatability risk to recover completed measurements without rerunning GPU work.",
"ordinal": 6,
"round_id": "4600ee4a-4b1e-404d-af1c-964ec63f3784",
"recorded_at": "2026-10-15T19:11:42.325151+00:00",
"approval_text": "yes approve",
"approval_source": "interactive campaign operator approval via Devin CLI",
"evidence_s3_url": "s3://pareton-s3/stage0/evidence/rounds/4600ee4a-4b1e-404d-af1c-964ec63f3784/4c218347-bab7-4499-b233-4f37134909aa-operator-recovery-20261015-20d5ca21b88d.tar.gz",
"measurements_changed": false,
"normal_drift_ceiling": 0.1,
"original_void_reason": "baseline_drift",
"actual_baseline_drift": 0.23783473312862324,
"frontend_code_changed": false,
"original_entry_reason": null,
"original_entry_status": "running",
"scoring_policy_change": "Apply #201 natural-stop v5 variance handling to all entries; retain raw variance and all correctness verdicts.",
"evidence_bundle_sha256": "sha256:f2bfb3957221bbf09a75039ad2b91a3c385e213e2bd43d2d0f0b0cda5434b64c",
"original_report_sha256": "805c244246a4e20092e2d2e475a752797cc5f8d18ab6ebcb82ac8f34d606f2d7",
"approved_preview_sha256": "20d5ca21b88d289dc5a377bb15ddea558eef3ed678abae1ab6878b72e6d3fafc",
"scoring_policy_revision": "33c2cca21dfa7c40691e94d265752865cf3cd194",
"global_threshold_changed": false,
"cancelled_successor_round_id": "489009d7-eb45-40a9-93d3-024f0873da12",
"approved_proposed_report_sha256": "d116269f2b8fa119214d02e94b23a8cc1362394d82fbeefbda9933dddafdc2c8",
"baseline_drift_waived_for_this_round_only": true
}
}