Skip to content

Actions: Leonccaa/ReVolta

Actions

All workflows

Actions

Loading...
Loading

Showing runs from all workflows
105 workflow runs
105 workflow runs

Filter by Workflow

⌥ + click/return to exclude

Filter by Event

⌥ + click/return to exclude

Filter by Status

⌥ + click/return to exclude

Filter by Branch

⌥ + click/return to exclude

Filter by Actor

⌥ + click/return to exclude
[Bugfix][Compile] Keep clones that change layout in the clone elimination pass checks #105: Pull request #95 reopened by Leonccaa
1m 44s fix/0.2-clone-cleanup-keep-layout-main
fix/0.2-clone-cleanup-keep-layout-main
1m 44s
[Metrics][Spec Decode] Report rejection-sampler residual fallbacks in /metrics checks #104: Pull request #89 reopened by Leonccaa
1m 6s f/0.2-rejection-fallback-metrics-main
f/0.2-rejection-fallback-metrics-main
1m 6s
[Core] Use the global CUDA graph pool for the Eagle/MTP draft graphs checks #103: Pull request #83 reopened by Leonccaa
1m 34s f/0.2-mtp-shared-graph-pool-main
f/0.2-mtp-shared-graph-pool-main
1m 34s
[Feature][KV Offload] Count CPU offload tier evictions by reason checks #102: Pull request #94 opened by Leonccaa
1m 35s f/0.2-host-l2-eviction-reasons-main
f/0.2-host-l2-eviction-reasons-main
1m 35s
[Bugfix][Frontend] Return output without logprobs when rows do not match tokens checks #101: Pull request #93 opened by Leonccaa
1m 20s fix/0.2-logprobs-row-mismatch-main
fix/0.2-logprobs-row-mismatch-main
1m 20s
[Bugfix][Compile] Keep returned clones that would alias graph inputs or outputs checks #100: Pull request #92 opened by Leonccaa
1m 40s fix/0.2-clone-cleanup-io-alias-main
fix/0.2-clone-cleanup-io-alias-main
1m 40s
[Perf][Core] Freeze the GPU worker GC heap after warmup checks #99: Pull request #91 opened by Leonccaa
1m 30s f/0.2-worker-gc-freeze-main
f/0.2-worker-gc-freeze-main
1m 30s
[Bugfix][Qwen4Exp] Use int64 row offsets in the HC and PLE Triton kernels checks #98: Pull request #90 opened by Leonccaa
1m 20s fix/0.2-hc-ple-row-offset-int64-main
fix/0.2-hc-ple-row-offset-int64-main
1m 20s
[Bugfix][Spec Decode] Keep the rejection residual in FP32 with a stable logsumexp checks #97: Pull request #88 opened by Leonccaa
1m 27s fix/0.2-rejection-residual-precision-main
fix/0.2-rejection-residual-precision-main
1m 27s
[Bugfix] Fail fast on incomplete local sharded safetensors checkpoints checks #96: Pull request #87 opened by Leonccaa
1m 40s fix/0.2-missing-shard-early-error-main
fix/0.2-missing-shard-early-error-main
1m 40s
[Bugfix] Hash zero multimodal limits into the model compile-cache key checks #95: Pull request #86 opened by Leonccaa
1m 37s fix/0.2-mm-limit-compute-hash-main
fix/0.2-mm-limit-compute-hash-main
1m 37s
[Bugfix] Bound the pending window of incremental detokenization checks #94: Pull request #85 opened by Leonccaa
1m 21s fix/0.2-detok-pending-cap-main
fix/0.2-detok-pending-cap-main
1m 21s
[Core] Reclaim target-loading temporaries before the draft model loads checks #93: Pull request #84 opened by Leonccaa
1m 33s f/0.2-reclaim-before-draft-load-main
f/0.2-reclaim-before-draft-load-main
1m 33s
[Core] Share one CUDA graph capture stream across the target and Eagle/MTP graphs checks #92: Pull request #82 opened by Leonccaa
1m 36s f/0.2-shared-capture-stream-main
f/0.2-shared-capture-stream-main
1m 36s
[Perf] Use blocking copy events in the MRv2 output and draft-token paths checks #91: Pull request #81 opened by Leonccaa
1m 38s f/0.2-mrv2-blocking-copy-event-main
f/0.2-mrv2-blocking-copy-event-main
1m 38s
[Bugfix][SM70] Pack the MTP router batch weight only for the 512x2560 router checks #90: Pull request #80 opened by Leonccaa
1m 12s f/0.2-mtp-router-batch-shape
f/0.2-mtp-router-batch-shape
1m 12s
[Perf][SM70] Pick the GDN input skinny GEMM column tiling by M (opt-in) checks #89: Pull request #79 opened by Leonccaa
1m 45s f/0.2-skinny-mplan
f/0.2-skinny-mplan
1m 45s
[Perf][QSA] SM70: launch grouped page4 groups longest first (opt-in) checks #88: Pull request #78 opened by Leonccaa
1m 19s f/0.2-qsa-grouped-lpt
f/0.2-qsa-grouped-lpt
1m 19s
[Perf][GDN] Build spec-decode query offsets on the host (opt-in) checks #87: Pull request #77 opened by Leonccaa
1m 17s f/0.2-gdn-cpu-spec-qsl
f/0.2-gdn-cpu-spec-qsl
1m 17s
[Perf][Core] Skip the all_token_ids copy for the V2 model runner checks #86: Pull request #76 opened by Leonccaa
1m 41s fix/0.2-mrv2-sched-no-token-copy-main
fix/0.2-mrv2-sched-no-token-copy-main
1m 41s
[Bugfix][KV offload] Keep external prefix hits on the Mamba page grid checks #85: Pull request #75 opened by Leonccaa
1m 26s f/0.2-offload-mamba-page-align
f/0.2-offload-mamba-page-align
1m 26s
[Perf][Core] Port partial prefix-cache hits for hybrid Mamba models checks #84: Pull request #70 synchronize by Leonccaa
1m 55s f/0.2-partial-prefix-cache
f/0.2-partial-prefix-cache
1m 55s
[Bugfix][Spec Decode] Insert resampled tokens with one warp checks #83: Pull request #74 opened by Leonccaa
1m 23s f/0.2-resample-insert-single-warp
f/0.2-resample-insert-single-warp
1m 23s
[Bugfix][SM70] Reject AWQ MoE TP shards that split quantization groups checks #82: Pull request #68 synchronize by Leonccaa
1m 10s f/0.2-awq-moe-tp-group-check
f/0.2-awq-moe-tp-group-check
1m 10s
[Bugfix] kv offload: fence pending stores before zeroing new pages checks #81: Pull request #73 opened by Leonccaa
1m 12s f/0.2-offload-zeroing-fence-main
f/0.2-offload-zeroing-fence-main
1m 12s