11月12月1月2月3月4月5月6月7月8月9月10月
周一周三周五
一年内 255863 次贡献
更少更多
- frostbyte_neo 推送了仓库 github-featured/affaan-m--everything-claude-code 的 main 分支比较 2 提交 »ca185ef5f7 chore(release): prepare signed 2.2.1 patch (#2920)a104765bf2 docs(ito-compute): document ito accept and ito_accept MCP workflow (#2893)
- frostbyte_neo 推送了仓库 github-featured/affaan-m--everything-claude-code 的 dependabot/npm_and_yarn/npm-minor-and-patch-e20e6656b6 分支比较 3 提交 »eff492cce7 chore(deps): bump the npm-minor-and-patch group across 1 directory with 6 updatesca185ef5f7 chore(release): prepare signed 2.2.1 patch (#2920)a104765bf2 docs(ito-compute): document ito accept and ito_accept MCP workflow (#2893)
- frostbyte_neo 推送了仓库 github-featured/run-llama--liteparse 的 main 分支比较 60 提交 »9a2caa475d Merge pull request #443 from run-llama/logan/fix-table-headersde8dad898a fix: try to properly absorb table headers16c311ea0f Merge pull request #442 from run-llama/logan/native-blocks-apibe7e82bc0c nit82f3657104 SpanCell carries an optional bbox, symmetric with Cell
- frostbyte_neo 推送了仓库 github-featured/lightseekorg--tokenspeed 的 unified-decode-path 分支比较 19 提交 »040c461ca5 fix(runtime): pass page table in forward metadata testsd233d9a729 fix(runtime): repair decode buffer import and lintf94f25c0d5 Merge branch 'main' into unified-decode-pathd5566e83f9 perf(kimi3): port dynamic AttnRes stride handling to gfx1250 (#1352)32e810f71b perf(k3): fuse MLA prefill preparation on gfx950 (#1351)
- frostbyte_neo 推送了仓库 github-featured/lightseekorg--tokenspeed 的 main 分支比较 5 提交 »6eef260147 perf(amd): hardcode the gfx1250 prefill occupancy cutoff (#1357)1cfe8ce913 perf(amd): port standard-cache DSA scorer to gfx1250 (#1354)ebda75ef29 perf(amd): derive gfx1250 prefill occupancy cutoff from the device (#1337)d5566e83f9 perf(kimi3): port dynamic AttnRes stride handling to gfx1250 (#1352)32e810f71b perf(k3): fuse MLA prefill preparation on gfx950 (#1351)
- frostbyte_neo 推送了仓库 github-featured/lightseekorg--tokenspeed 的 shared/glm-5.3-flash 分支比较 26 提交 »bb04864381 test: preserve CuTe compile option binding9d85a2748b ci: deselect known MI450 simulator failures899624d13a refactor(attention): route DSA dense backend through registry8459a3cdc9 refactor(glm-5.3-flash): simplify attention integrationac19d29493 fix: select MLA backend inside AMD DSA
- frostbyte_neo 推送了仓库 github-featured/lightseekorg--tokenspeed 的 hongtaoc/flashinfer-0618-kda-decode 分支比较 2 提交 »ef96140d3f Fuse KDA decode state stagingd531e67e84 Preserve FlashInfer KDA decode across L2 restore
- frostbyte_neo 推送了仓库 github-featured/lightseekorg--tokenspeed 的 feat/dflash2-official-node-cudagraph 分支比较 117 提交 »48cb186b0b feat: add DFlash2 draft serving support9d50421a89 perf(gemm): 256-bit skinny GEMV configs for the measured decode shapes (#1349)3898df1054 fix(profile): make the CUPTI graph-capture warm-up opt-in (#1348)9104ebb6a0 Update tokenspeed-smg* pins to 1.10.1.post20260901 / 0.4.16.post20260901 / 0.9.1.post20260901 (#1344)bd56852b5c ci: fix B300 DeepSWE runtime access (#1317)
- frostbyte_neo 推送了仓库 github-featured/tashfeenahmed--freellmapi 的 main 分支比较 10 提交 »4a8f0950fe Prune the quota observation log in 5k-row chunks (#1124)77e0ecc241 perf: stop the quota panel from ranking the whole observation log on every poll (#1123)d8fae972b9 feat(desktop): add an rpm target to the Linux desktop build (#981)c0bd84aee7 feat(analytics): sortable columns on the recent-calls, usage-by-key and per-model tables (#495)0ede822d0a feat(analytics): show the full error message on hover in the recent errors table (#489)
- frostbyte_neo 推送了仓库 github-featured/everyinc--compound-engineering… 的 tmchow/debug-issue-1607 分支比较 5 提交 »5c26b31f14 fix(cross-model-review): state the fold-in read as one conditiona58d268bbd fix(peer-job-runner): reserve exit 3 for a genuinely absent artifacte4e9c326d2 test(peer-job-runner): exercise the ownership branch, not mode bits7f6c3b51d8 docs(solutions): capture the expected-absence-vs-error CLI lesson2eda232284 fix(peer-job-runner): name the outcome when a peer artifact is absent
- frostbyte_neo 推送了仓库 github-featured/everyinc--compound-engineering… 的 tmchow/fix-babysit-not-detecting 分支比较 3 提交 »bea9dad4a0 docs(skill-design): a gate's inputs must be read where the gate runs9b1c5cbbd4 perf(ce-babysit-pr): stop spending agent tokens on speculative wakes58164ee99e fix(ce-commit-push-pr): read auto_babysit at the handoff that consumes it