12月1月2月3月4月5月6月7月8月9月10月11月
周一周三周五
一年内 264198 次贡献
更少更多
- frostbyte_neo 推送了仓库 github-featured/smol-machines--smolvm 的 cuda-pool-event-reconcile 分支2eb7a39ee0 Prepare CUDA fork handoffs in parallel (#816)
- frostbyte_neo 推送了仓库 github-featured/smol-machines--smolvm 的 cuda-pool-activation-retry 分支比较 3 提交 »dfbc900efc Start fork lease TTL after activatione9702ce078 Make held-fork activation retries idempotent34222a5fbd Retry transient fork lease payload staging
- frostbyte_neo 推送了仓库 github-featured/smol-machines--smolvm 的 cuda-pool-activation-refill 分支97e72540fc Defer pool refills until activation completes
- frostbyte_neo 推送了仓库 github-featured/smol-machines--smolvm 的 cuda-path3-density 分支比较 3 提交 »4efe77df67 Path 3 density: verify shared-chunk content against upload CRCs at fork time so kernel-written chunks auto-degrade to private7e17a8204b Path 3: opt-in weight-sharing density — share fully-H2D-covered VMM chunks across clone workersa5704ef720 Path 3: route TCP clone connections to worker processes so remote clients can fork against a network GPU host
- frostbyte_neo 推送了仓库 github-featured/smol-machines--smolvm 的 cuda-path3 分支比较 4 提交 »72ef494eb5 Path 3: detect the clone-worker socket family via getsockname (SO_DOMAIN is Linux-only and broke the macOS build)cf71da66a9 Path 3 density: verify shared-chunk content against upload CRCs at fork time so kernel-written chunks auto-degrade to private0c29dc9bc5 Path 3: opt-in weight-sharing density — share fully-H2D-covered VMM chunks across clone workers9311321d15 Path 3: route TCP clone connections to worker processes so remote clients can fork against a network GPU host
- frostbyte_neo 推送了仓库 github-featured/smol-machines--smolvm 的 cuda-parallel-fork-handoff 分支e43f1065b2 Retry transient KVM failures for fork clones (#817)
- frostbyte_neo 推送了仓库 github-featured/smol-machines--smolvm 的 cuda-module-stream-handoff 分支03efe3d639 Map CUDA host snapshots directly across clone workers (#813)
- frostbyte_neo 推送了仓库 github-featured/smol-machines--smolvm 的 cuda-module-mmap-mainline 分支64ea7afbc2 Stream CUDA module handoffs from shared images (#812)