sskj/deploy/profiles/pro6000/glm53_nvfp4_pro6000_sglang_tp8dflash2.env
yy-fighting 5c749cda03 feat(pro6000/GLM-5.3): 方案 D/E/F 部署资产入库(TP2PP4 生产配方 / TP8+DFlash2 / PD 分离四角色链)
- scripts: 11 个服务器原样脚本入库(md5 对照表更新至 README);D=60.1 生产原样配方、
  E=v5 DFlash 底稿、F=PD 链四角色部署+launch+双场景压测驱动
- profiles: 新增 6 个 .env(D/E 单机 + F 四角色,均带镜像 digest
  sha256:28e0d260…,对齐 kimi3 PD 多角色先例)
- deploy/PD_CHAIN.md: 方案 F 编排手册(启动顺序 mc-master→prefill→decode→router、
  基础设施依赖表、质量门口径、拆链恢复、÷2 单机等效判决)
- platforms/patches/pro6000/glm53_pd_chain/: sglang 补丁树 vs 镜像原版 11 文件
  unified diff 快照——宿主树无 .git,此为唯一版本记录(DFlash+PP+PD 解锁全集)
- deploy/manifests/: GLM-5.3-NVFP4(47分片)/GLM-5.3-DFlash2(单分片) 权重 md5 清单
- deploy/CURRENT.md: 全集群现役状态页(2026-09-08 八机实测)
- deploy/verify_profile.sh: 防漂移核验工具(digest+参数 token 比对+端口/health,
  已在 60.1 生产容器实测 PASS)
2026-09-08 16:36:15 +08:00

53 lines
3.2 KiB
Bash
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

# GLM-5.3-NVFP4 SGLang TP=8 + DFlash2 speculative decoding profile (single RTX 6000D node, 8 GPUs).
# 方案 E2026-09-08 场景一补测6000D-2
# 可执行部署脚本experiments/pro6000/glm53_nvfp4_pro6000d_sglang_dual_scenario_bench/scripts/deploy_glm53_tp8_dflash2.sh
# v5 终版底稿,含四轮 OOM 战役完整教训注释md5 5bf2b47c9349e5855b963e571e35c096
#
# 关键点(实测踩坑,勿随意改):
# - v5 配方核心MRR 48→12verify CUDA graph 4.04→0.83GB,真正起作用的杠杆)+
# --speculative-draft-window-size 2048 + mem0.85/cps8192。四轮 OOM 根因与推导见脚本头注释
# - DFLASH block-diffusion 草稿 7 tokens/步draft 权重 GLM-5.3-DFlash2fa4 draft
# attentionfa4 会把 draft KV 强制 bf16fp8 需换 flashinfer/triton 后端,仅省 0.35GB 未用)
# - KV 池 243,584 tokenkv fp8_e4m3 由模型配置自动带出(无需显式 flag
# - 底稿(本 profile LAUNCH_ARGSradix/AR 均为禁用;场景一实测变体共四处 delta
# ① 去 --disable-radix-cache90% 命中前提)② 去 --disable-custom-all-reduce
# v1 CAR 与方案 A 一致开启)③ 加 --context-length 270336 ④ 加 --reasoning-parser
# glm45 --tool-call-parser glm47质量门 7/7 的前提)
# - 判决:场景一 8 点全部低于方案 C、7 点低于 A——DFlash accept 低于 EAGLE同语料
# 2.53 vs 2.84)而每步墙钟相当,劣势全在接受率。投机栈选型维持 EAGLE3勿用
# DFlash2 替换(性能问题非质量问题)
#
# 实测成绩飞书《GLM-5.3-NVFP4 双场景压测报告》方案 E 行(文档 SZUSdEqY1oRVxGxgILBcHqPJnEc
# 质量:变体配置下质量门 7/7GSM8K×5、中文推理、tool call 全过DFlash 草稿无质量损失。
PLATFORM=pro6000
EXPERIMENT=glm53_nvfp4_pro6000_sglang_tp8dflash2
MODEL_NAME=GLM-5.3-NVFP4
ENGINE=sglang
RUNTIME=docker
DOCKER_IMAGE=lmsysorg/sglang:nightly-dev-20260828-daf63171
DOCKER_IMAGE_DIGEST=sha256:28e0d26073161e49ca56eba808d264a4223804a212020f1dfe1b2405b9f8a399
CONTAINER_NAME=glm53-tp8-dflash2
MODEL_PATH=/data/hf_models/GLM-5.3-NVFP4
DRAFT_MODEL_PATH=/data/hf_models/GLM-5.3-DFlash2
SERVED_MODEL_NAME=/data/hf_models/GLM-5.3-NVFP4
PORT=30000
HEALTH_PATH=/health
HEALTH_WAIT_S=900
CONTAINER_PYTHON=python3
TP=8
MEM_FRACTION_STATIC=0.85
MAX_RUNNING_REQUESTS=12
CHUNKED_PREFILL_SIZE=8192
DEVICE_VARS="CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7"
ENGINE_ENV="PYTHONUNBUFFERED=1 HF_HUB_OFFLINE=1 TRANSFORMERS_OFFLINE=1"
DOCKER_FLAGS="--gpus all --shm-size 64g --ipc=host -p ${PORT}:${PORT}"
VOLUMES="/data/hf_models:/data/hf_models"
BOOTSTRAP="python3 -m sglang.launch_server ${LAUNCH_ARGS}"
LAUNCH_ARGS="--model-path ${MODEL_PATH} --tp-size ${TP} --mem-fraction-static ${MEM_FRACTION_STATIC} --max-running-requests ${MAX_RUNNING_REQUESTS} --disable-radix-cache --disable-shared-experts-fusion --moe-runner-backend flashinfer_cutlass --disable-flashinfer-autotune --disable-custom-all-reduce --chunked-prefill-size ${CHUNKED_PREFILL_SIZE} --speculative-algorithm DFLASH --speculative-draft-model-path ${DRAFT_MODEL_PATH} --speculative-draft-attention-backend fa4 --speculative-draft-window-size 2048 --host 0.0.0.0 --port ${PORT} --json-model-override-args {\"index_topk_freq\": 4}"