shishi 455a78161b fix(910c/glm52): 对齐官方A3教程参数(修DP die分配问题,同dsv4 99a22f0) ...

dsv4实验发现DP副本绑定到同一组die的问题(99a22f0),glm52存在相同问题:
缺--max-num-batched-tokens和--api-server-count导致DP worker设备分配异常
(不加api-server-count时vllm为N个DP rank启动N个API server)。

对齐docs.vllm.ai GLM5.2 A3官方教程:
- 加 --max-num-batched-tokens 8192 (官方值,影响DP调度)
- 加 --api-server-count 1 (官方值,避免多API server干扰设备分配)
- 去掉 --kv-cache-dtype fp8 (官方不指定,用默认bfloat16;且此镜像fp8本就未生效)
- 保留 --trust-remote-code / --enable-expert-parallel / enable_dsa_cp (官方有)

2026-07-29 15:28:54 +08:00

..

adaptive_config.env

feat(platform): add Ascend 910C NPU platform support

2026-07-27 22:00:05 +08:00

compare.py

feat(platform): add Ascend 910C NPU platform support

2026-07-27 22:00:05 +08:00

config.env

fix(910c/glm52): 对齐官方A3教程参数(修DP die分配问题,同dsv4 99a22f0)

2026-07-29 15:28:54 +08:00

generate_scenarios.py

feat(platform): add Ascend 910C NPU platform support

2026-07-27 22:00:05 +08:00

matrix.json

feat(platform): add Ascend 910C NPU platform support

2026-07-27 22:00:05 +08:00

run_adaptive_concurrency_add16.sh

fix(910c/glm52): 对齐官方A3教程参数(修DP die分配问题,同dsv4 99a22f0)

2026-07-29 15:28:54 +08:00

run_adaptive_concurrency.sh

feat(platform): add Ascend 910C NPU platform support

2026-07-27 22:00:05 +08:00

run_bench.sh

feat(platform): add Ascend 910C NPU platform support

2026-07-27 22:00:05 +08:00

start_vllm_docker.sh

fix(910c/glm52): 对齐官方A3教程参数(修DP die分配问题,同dsv4 99a22f0)

2026-07-29 15:28:54 +08:00

start_vllm_dp.sh

feat(platform): add Ascend 910C NPU platform support

2026-07-27 22:00:05 +08:00

TP8_vs_TP16_report.md

fix(910c/glm52): 修复sglang0.5.2解析兼容性+TP=16设备挂载+health超时

2026-07-28 16:43:05 +08:00