Skip to content

Actions: huawei-csl/KVarN

Actions

pre-commit

Actions

Loading...
Loading

Show workflow options

Create status badge

Loading
pre-commit.yml
104 workflow runs
104 workflow runs

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
KVarN: port to GLM-5.2 sparse-MLA (DSA) backend pre-commit #105: Commit 7586257 pushed by philippebich
5m 22s main
main
5m 22s
KVarN: size the split-K partial buffer by sm_count, not max_num_seqs pre-commit #104: Commit 115437d pushed by philippebich
5m 1s main
main
5m 1s
docs: update vLLM version to 0.23.0 in README pre-commit #103: Commit 39865ad pushed by philippebich
8m 15s main
main
8m 15s
Merge vllm-0.23 into main pre-commit #102: Commit a123370 pushed by philippebich
5m 12s main
main
5m 12s
KVarN: support non-causal attention for DFlash speculative decoding pre-commit #101: Commit f78be57 pushed by philippebich
5m 19s main
main
5m 19s
KVarN: trust the weight manifest only when all listed shards exist pre-commit #100: Commit a601d2a pushed by philippebich
8m 27s main
main
8m 27s
Merge pull request #20 from sztlink/fix/weight-bytes-double-count pre-commit #99: Commit 4f403bd pushed by philippebich
5m 46s main
main
5m 46s
KVarN: CUDA graph capture for the speculative verify pass pre-commit #98: Commit e04787c pushed by philippebich
4m 5s main
main
4m 5s
fix(kvarn): avoid double-counting weights when a repo ships both consolidated and sharded safetensors pre-commit #97: Pull request #20 opened by sztlink
7s sztlink:fix/weight-bytes-double-count
sztlink:fix/weight-bytes-double-count
7s
docs: Running on WSL2 section (residency limit, spill diagnostic, UVA note) pre-commit #96: Pull request #19 synchronize by sztlink
7s sztlink:wsl-readme-note
sztlink:wsl-readme-note
7s
KVarN: fused multi-query verify kernel for speculative decoding pre-commit #95: Commit 98f9a2e pushed by philippebich
5m 27s main
main
5m 27s
KVarN: fix dense-path KV corruption under prefix caching (issue #10 l… pre-commit #94: Commit 0af14d6 pushed by philippebich
6m 18s main
main
6m 18s
KVarN-MLA: support prefix caching and chunked prefill pre-commit #93: Commit 2e8f5ca pushed by philippebich
5m 40s main
main
5m 40s
KVarN: charge lazy state to the memory profile, right-size split-K bu… pre-commit #92: Commit e9428e9 pushed by philippebich
5m 54s main
main
5m 54s
KVarN: prefer the preset's tile size as the kv-cache block size pre-commit #91: Commit 1311917 pushed by philippebich
5m 19s main
main
5m 19s
KVarN: batched spec-verify attention + decode-kernel warmup (issue #10) pre-commit #90: Commit 12241b6 pushed by philippebich
4m 32s main
main
4m 32s
docs: Running on WSL2 section (residency limit, spill diagnostic, UVA note) pre-commit #89: Pull request #19 opened by sztlink
10s sztlink:wsl-readme-note
sztlink:wsl-readme-note
10s
docs: README note now lists supported page sizes 64 and 128 pre-commit #88: Commit 002d388 pushed by philippebich
5m 32s main
main
5m 32s
KVarN: support tile / block size 64 in addition to 128 pre-commit #87: Commit 42899a5 pushed by philippebich
5m 46s main
main
5m 46s
docs: add README sections for hybrid models and speculative decoding pre-commit #86: Commit bb71da2 pushed by philippebich
5m 56s main
main
5m 56s
KVarN: fix garbage output with speculative decoding (MTP / draft) pre-commit #85: Commit d7bbff1 pushed by philippebich
5m 33s main
main
5m 33s
KVarN: Mamba-aware pool budget — size the fp16 pool by attention laye… pre-commit #84: Commit 46d0b21 pushed by philippebich
5m 46s main
main
5m 46s
Merge pull request #16 from sztlink/wsl-throughput-fix pre-commit #83: Commit 232e701 pushed by philippebich
5m 28s main
main
5m 28s
KVarN: default Sinkhorn iterations 16 -> 8 (lossless, ~halves flush c… pre-commit #81: Commit 343fe14 pushed by philippebich
5m 15s main
main
5m 15s
README update pre-commit #80: Commit d784952 pushed by philippebich
10m 33s main
main
10m 33s