- 31 Oct, 2025 6 commits
- 29 Oct, 2025 2 commits
- 27 Oct, 2025 2 commits
- 24 Oct, 2025 4 commits
- 23 Oct, 2025 2 commits
- 20 Oct, 2025 2 commits
- 17 Oct, 2025 2 commits
- 16 Oct, 2025 3 commits
- 15 Oct, 2025 6 commits
- 13 Oct, 2025 11 commits
增加pd分离单实例跨机第二个ip通过配置文件获取。配置文件上设置如下: See merge request dcutoolkit/deeplearing/vllm!234
# 第一个ip为D的第一个节点,第二个ip为D的第二个节点,配置:export IP_CONFIG_FILE=/data/xiabo/w4a8_1/ip_config.txt 192.168.1.1 192.168.1.100 192.168.1.2 192.168.1.101 192.168.1.3 192.168.1.102 10.16.1.75 10.16.1.76
add VLLM_USE_LIGHTOP_MOE_SUM_MUL_ADD
set USE_FUSED_RMS_QUANT=1 and USE_FUSED_SILU_MUL_QUANT=1
fix the error in test_moe caused by moe align not supporting 511 multi-modal switching to torch implementation on z100l&k100
修复awq 的mtp中的blockint8的问题 See merge request dcutoolkit/deeplearing/vllm!229
删除DPSK_FP16_QUICK,以及增加awq和blockwiseint8的shared_output接口 See merge request dcutoolkit/deeplearing/vllm!228
去掉all2all ep相关代码 See merge request dcutoolkit/deeplearing/vllm!226