2.2.0

ascend-robot[pytorch][bugfix]fix hunyuan tune bug

dda31b2c创建于 2025年10月16日历史提交

文件	最后提交记录	最后更新时间
chat_qwen25_7b_lora_ptd.sh	!2325 添加Qwen2.5 0.5B/1.5B/3B LoRA微调,权重合并及chat脚本 Qwen2.5其他微调脚本添加padded-samples参数 Merge pull request !2325 from 商元义/master	1 年前
chat_qwen25_7b_ptd.sh	!2702 Added qwen25_7b conversation scripts Merge pull request !2702 from guozhihua/master	1 年前
ckpt_convert_qwen25_hf2mcore.sh	!2828 [pytorch][sh]Add CUDA_DEVICE_MAX_CONNECTIONS Merge pull request !2828 from yanzhixiao/add-environment-variable	11 个月前
ckpt_convert_qwen25_mcore2hf.sh	!2803 fix readme of V3 Merge pull request !2803 from qu_yueze/master	11 个月前
ckpt_convert_qwen25_mcore2hf_lora.sh	!2803 fix readme of V3 Merge pull request !2803 from qu_yueze/master	11 个月前
data_convert_qwen25_instruction.sh	!2803 fix readme of V3 Merge pull request !2803 from qu_yueze/master	11 个月前
data_convert_qwen25_instruction_history.sh	!2342 Qwen2.5-7B问题单修复 Merge pull request !2342 from changlei/master_dts_jpj	1 年前
data_convert_qwen25_instruction_pack.sh	!2285 增加Qwen2.5-3B|7B|14B|32B和Qwen2-MoE微调脚本 Merge pull request !2285 from changlei/master	1 年前
data_convert_qwen25_pretrain.sh	!1720 add qwen2.5-7b Merge pull request !1720 from LeiZhenzhen/master	1 年前
evaluate_qwen25_0point5b_ptd.sh	!2104 phi3.5-moe 补充ut用例 Merge pull request !2104 from changlei/master	1 年前
evaluate_qwen25_14b_ptd.sh	!1771 适配Qwen2.5-14B Merge pull request !1771 from changlei/master	1 年前
evaluate_qwen25_1point5b_ptd.sh	!1793 添加Qwen2.5-1.5B模型 Merge pull request !1793 from caoruichao/master	1 年前
evaluate_qwen25_32b_ptd.sh	!1793 添加Qwen2.5-1.5B模型 Merge pull request !1793 from caoruichao/master	1 年前
evaluate_qwen25_3b_ptd.sh	!1740 适配Qwen2.5-3B Merge pull request !1740 from changlei/master	1 年前
evaluate_qwen25_72b_ptd.sh	!1904 添加Qwen2.5-72B模型 Merge pull request !1904 from 商元义/qwen25_0point5b	1 年前
evaluate_qwen25_7b_ptd.sh	!1720 add qwen2.5-7b Merge pull request !1720 from LeiZhenzhen/master	1 年前
generate_qwen25_0point5b_ptd.sh	!1901 添加Qwen2.5-0.5B适配 Merge pull request !1901 from 商元义/qwen25_0point5b	1 年前
generate_qwen25_14b_ptd.sh	!1771 适配Qwen2.5-14B Merge pull request !1771 from changlei/master	1 年前
generate_qwen25_1point5b_ptd.sh	!1793 添加Qwen2.5-1.5B模型 Merge pull request !1793 from caoruichao/master	1 年前
generate_qwen25_32b_ptd.sh	!1793 添加Qwen2.5-1.5B模型 Merge pull request !1793 from caoruichao/master	1 年前
generate_qwen25_3b_ptd.sh	!1740 适配Qwen2.5-3B Merge pull request !1740 from changlei/master	1 年前
generate_qwen25_72b_ptd.sh	!1904 添加Qwen2.5-72B模型 Merge pull request !1904 from 商元义/qwen25_0point5b	1 年前
generate_qwen25_7b_lora_ptd.sh	!1819 专家以及共享专家支持tp+ep切分&MiniCPM3预训练性能修复&Qwen-LoRA推理bug修复 Merge pull request !1819 from 商元义/master	1 年前
generate_qwen25_7b_ptd.sh	!1720 add qwen2.5-7b Merge pull request !1720 from LeiZhenzhen/master	1 年前
pretrain_qwen25_0point5_32k_ptd.sh	!2507 Rename all seq length naming K to k Merge pull request !2507 from shenjiarun/master	1 年前
pretrain_qwen25_14b_32k_ptd.sh	!2836 [pytorch][sh]update Qwen2.5 14b/72b 32K scripts Merge pull request !2836 from sunjunjie/master	11 个月前
pretrain_qwen25_14b_8k_pack_ptd.sh	!2185 Qwen2.5性能调优 Merge pull request !2185 from zengshu/master	1 年前
pretrain_qwen25_1point5b_32k_ptd.sh	!2507 Rename all seq length naming K to k Merge pull request !2507 from shenjiarun/master	1 年前
pretrain_qwen25_32b_32k_ptd.sh	!2582 fix_adaptor mtp Merge pull request !2582 from jzh/fix_adaptor	1 年前
pretrain_qwen25_3b_32k_ptd.sh	!2507 Rename all seq length naming K to k Merge pull request !2507 from shenjiarun/master	1 年前
pretrain_qwen25_72b_32k_ptd.sh	!2836 [pytorch][sh]update Qwen2.5 14b/72b 32K scripts Merge pull request !2836 from sunjunjie/master	11 个月前
pretrain_qwen25_72b_4k_pack.sh	[pytorch][bugfix]fix hunyuan tune bug Co-authored-by: guozhihua<guozhihua2@huawei.com> # message auto-generated for no-merge-commit merge: !3519 merge huanyuan_tune_fix_2.2 into 2.2.0 [pytorch][bugfix]fix hunyuan tune bug Created-by: guozhihua2 Commit-by: guozhihua Merged-by: ascend-robot Description: 1.change hunyuan tune pad-to-multiple-of to 8 and add use-flash-attn 2.add recompute-activation-function in qwen25-72b-4k-pack See merge request: Ascend/MindSpeed-LLM!3519	7 个月前
pretrain_qwen25_72b_4k_pack_A3.sh	[pytorch][bugfix]fix hunyuan tune bug Co-authored-by: guozhihua<guozhihua2@huawei.com> # message auto-generated for no-merge-commit merge: !3519 merge huanyuan_tune_fix_2.2 into 2.2.0 [pytorch][bugfix]fix hunyuan tune bug Created-by: guozhihua2 Commit-by: guozhihua Merged-by: ascend-robot Description: 1.change hunyuan tune pad-to-multiple-of to 8 and add use-flash-attn 2.add recompute-activation-function in qwen25-72b-4k-pack See merge request: Ascend/MindSpeed-LLM!3519	7 个月前
pretrain_qwen25_72b_8k_pack_ptd.sh	!2761 Optimize qwen2.5 72b Merge pull request !2761 from sunjunjie/qwen25	11 个月前
pretrain_qwen25_7b_32k_ptd.sh	!2507 Rename all seq length naming K to k Merge pull request !2507 from shenjiarun/master	1 年前
pretrain_qwen25_7b_8k_pack_ptd.sh	!2185 Qwen2.5性能调优 Merge pull request !2185 from zengshu/master	1 年前
tune_qwen25_0point5b_4k_full_pack.sh	!2323 添加Qwen2.5 pack模式0.5B/1.5B/72B pack模式微调脚本 Merge pull request !2323 from 商元义/qwen_tune_pack	1 年前
tune_qwen25_0point5b_4k_full_ptd.sh	!2346 添加Qwen2.5 0.5B/1.5B/3B微调脚本 Merge pull request !2346 from CY-Slightwind/qwen25	1 年前
tune_qwen25_0point5b_4k_lora_ptd.sh	!2325 添加Qwen2.5 0.5B/1.5B/3B LoRA微调,权重合并及chat脚本 Qwen2.5其他微调脚本添加padded-samples参数 Merge pull request !2325 from 商元义/master	1 年前
tune_qwen25_14b_4k_full_pack.sh	!2285 增加Qwen2.5-3B|7B|14B|32B和Qwen2-MoE微调脚本 Merge pull request !2285 from changlei/master	1 年前
tune_qwen25_14b_4k_full_ptd.sh	!2325 添加Qwen2.5 0.5B/1.5B/3B LoRA微调,权重合并及chat脚本 Qwen2.5其他微调脚本添加padded-samples参数 Merge pull request !2325 from 商元义/master	1 年前
tune_qwen25_14b_4k_lora_pack.sh	!2817 [pytorch][bugfix]Limit the simultaneous use of CCLORA and overlap-param-gather Merge pull request !2817 from qu_yueze/master	11 个月前
tune_qwen25_14b_4k_lora_ptd.sh	!2817 [pytorch][bugfix]Limit the simultaneous use of CCLORA and overlap-param-gather Merge pull request !2817 from qu_yueze/master	11 个月前
tune_qwen25_1point5b_4k_full_pack.sh	!2323 添加Qwen2.5 pack模式0.5B/1.5B/72B pack模式微调脚本 Merge pull request !2323 from 商元义/qwen_tune_pack	1 年前
tune_qwen25_1point5b_4k_full_ptd.sh	!2346 添加Qwen2.5 0.5B/1.5B/3B微调脚本 Merge pull request !2346 from CY-Slightwind/qwen25	1 年前
tune_qwen25_1point5b_4k_lora_ptd.sh	!2325 添加Qwen2.5 0.5B/1.5B/3B LoRA微调,权重合并及chat脚本 Qwen2.5其他微调脚本添加padded-samples参数 Merge pull request !2325 from 商元义/master	1 年前
tune_qwen25_32b_16k_full_pack.sh	!2913 qwen25-32b-16k/32k-sft-pack脚本添加 Merge pull request !2913 from liuchangkun/master	10 个月前
tune_qwen25_32b_32k_full_pack.sh	!2913 qwen25-32b-16k/32k-sft-pack脚本添加 Merge pull request !2913 from liuchangkun/master	10 个月前
tune_qwen25_32b_4k_full_pack.sh	!2285 增加Qwen2.5-3B|7B|14B|32B和Qwen2-MoE微调脚本 Merge pull request !2285 from changlei/master	1 年前
tune_qwen25_32b_4k_full_ptd.sh	!2794 fix oom of tune of qwen25-32b Merge pull request !2794 from qu_yueze/master	11 个月前
tune_qwen25_32b_4k_lora_ptd.sh	!2817 [pytorch][bugfix]Limit the simultaneous use of CCLORA and overlap-param-gather Merge pull request !2817 from qu_yueze/master	11 个月前
tune_qwen25_3b_4k_full_pack.sh	!2285 增加Qwen2.5-3B|7B|14B|32B和Qwen2-MoE微调脚本 Merge pull request !2285 from changlei/master	1 年前
tune_qwen25_3b_4k_full_ptd.sh	!2346 添加Qwen2.5 0.5B/1.5B/3B微调脚本 Merge pull request !2346 from CY-Slightwind/qwen25	1 年前
tune_qwen25_3b_4k_lora_ptd.sh	!3154 [pytorch][feature]the parameter --use-mc2 has been renamed to --use-ascend-mc2 Merge pull request !3154 from yanzhixiao/update-args	9 个月前
tune_qwen25_72b_16k_full_pack.sh	!2812 add qwen25 16K pack full sft Merge pull request !2812 from chenzeng/master	11 个月前
tune_qwen25_72b_32k_full_pack.sh	!2900 qwen2.5 sft 72b-32k pack Merge pull request !2900 from 蒋鹏军/master	11 个月前
tune_qwen25_72b_4k_full_pack.sh	!2323 添加Qwen2.5 pack模式0.5B/1.5B/72B pack模式微调脚本 Merge pull request !2323 from 商元义/qwen_tune_pack	1 年前
tune_qwen25_72b_4k_full_ptd.sh	!2325 添加Qwen2.5 0.5B/1.5B/3B LoRA微调,权重合并及chat脚本 Qwen2.5其他微调脚本添加padded-samples参数 Merge pull request !2325 from 商元义/master	1 年前
tune_qwen25_72b_4k_lora_ptd.sh	!3252 [pytorch][bugfix]Fix LoRA finetune bug Merge pull request !3252 from sunjunjie/master	8 个月前
tune_qwen25_7b_4k_full_pack.sh	!2285 增加Qwen2.5-3B|7B|14B|32B和Qwen2-MoE微调脚本 Merge pull request !2285 from changlei/master	1 年前
tune_qwen25_7b_4k_full_ptd.sh	!2325 添加Qwen2.5 0.5B/1.5B/3B LoRA微调,权重合并及chat脚本 Qwen2.5其他微调脚本添加padded-samples参数 Merge pull request !2325 from 商元义/master	1 年前
tune_qwen25_7b_4k_lora_ptd.sh	!2817 [pytorch][bugfix]Limit the simultaneous use of CCLORA and overlap-param-gather Merge pull request !2817 from qu_yueze/master	11 个月前
tune_qwen25_7b_full_ptd.sh	!1933 添加Qwen2.5-LoRA微调，修改全参微调bug Merge pull request !1933 from 商元义/qwen25_0point5b	1 年前
tune_qwen25_7b_lora_ptd.sh	!2169 core0.8.0 upgrade to mindspeed12.25 Merge pull request !2169 from MeiFei/core0.8.0-1225	1 年前