• Joined on 2026-07-22
yy pushed to main at qqtang/sskj 2026-08-11 03:55:41 +00:00
fe375e1307 docs(pd): 补充运维一键部署完整步骤(干净环境从零到跑通)
yy pushed to main at qqtang/sskj 2026-08-11 03:42:52 +00:00
ddf807d458 feat(bench/pd): 支持 --flush-cache 透传 + PD profile 加 --disable-radix-cache
yy pushed to main at qqtang/sskj 2026-08-11 03:19:44 +00:00
da0e1b4372 fix(pd): config.env 加 DOCKER_CLIENT_IMAGE(bench 用 docker client 复用 kimi-k3 镜像)
yy pushed to main at qqtang/sskj 2026-08-11 02:50:46 +00:00
04dfa31583 fix(pd): PATCH_MOUNTS 补充 flashkda wheel 挂载(BOOTSTRAP 需要 /flash_kda-*.whl)
yy pushed to main at qqtang/sskj 2026-08-11 02:36:27 +00:00
d72dbff689 feat(pro6000): Kimi-K3 PD 分离部署(MoonCake RDMA)- 8 节点 P/D 双 profile + deploy_pd.sh 编排 + 文档
yy pushed to main at qqtang/sskj 2026-08-10 03:10:23 +00:00
3bd04698bb feat(pro6000): Kimi-K3 TP32×EP32 部署 profile、sm_120 补丁与运维手册 - 4 节点 RoCE 部署 + bench 实验
yy pushed to main at qqtang/sskj 2026-08-03 09:22:05 +00:00
d9a2e3b5e5 fix(910c/glm52): 镜像与 client 修正 - 部署在 910c.2 的 GLM5.2-tuned 镜像
yy pushed to main at qqtang/sskj 2026-08-03 08:41:09 +00:00
f33c5f1d3d fix(deploy): DP_FLAG 未达 BOOTSTRAP 启动命令 + bench 的 model/tokenizer 分离
yy pushed to main at qqtang/sskj 2026-08-03 07:43:25 +00:00
ea8302561e fix(910c): bench client 可用化 - 本地 vllm-ascend-sglang 镜像 + torch_npu 自动加载禁用
yy pushed to main at qqtang/sskj 2026-08-03 07:35:51 +00:00
6ba04325d3 feat(910c): 部署解耦 - vLLM-Ascend profile 与 deploy 层接管服务启停
yy pushed to main at qqtang/sskj 2026-08-03 07:24:54 +00:00
9acf9fdfdb feat(pro6000): 部署/测试解耦 - deploy 层支持多节点与 vLLM,新增 6 个 profile
yy pushed to main at qqtang/sskj 2026-08-02 07:37:11 +00:00
3761d75b00 feat(ops): add unified bench/deploy layers and P800 profile
yy pushed to main at qqtang/sskj 2026-07-30 02:48:11 +00:00
e885fd0dc2 feat(adaptive): support tiered per-ISL TTFT SLO via get_ttft_slo_ms()
yy pushed to main at qqtang/sskj 2026-07-30 02:05:35 +00:00
ad4fd2b878 fix: INVALID_WORKLOAD 只跳过当前shape而不中止所有配置
yy pushed to main at qqtang/sskj 2026-07-29 08:49:37 +00:00
4914ff4041 fix(dsv4): disable MTP speculative decoding for fair H20 comparison
yy pushed to main at qqtang/sskj 2026-07-29 07:29:13 +00:00
455a78161b fix(910c/glm52): 对齐官方A3教程参数(修DP die分配问题,同dsv4 99a22f0)
yy pushed to main at qqtang/sskj 2026-07-29 07:08:49 +00:00
99a22f05b8 fix(dsv4): align launch params with official A3 tutorial (fixes DP die allocation)
yy pushed to main at qqtang/sskj 2026-07-29 05:44:51 +00:00
c222ed98b2 feat(910c/glm52): 并行配置改为 4 4 / 8 2 / 16 1 对标H20的 2 4 / 4 2 / 8 1
yy pushed to main at qqtang/sskj 2026-07-29 05:36:50 +00:00
a65849b77d fix(dsv4): use all 16 dies with TP4/DP4 + TP8/DP2 + TP16/DP1
yy pushed to main at qqtang/sskj 2026-07-29 03:52:13 +00:00
63ab41b65a docs: consolidate project docs (dedup, relocate, expand 910C client guide)