2.2.0

ascend-robot[pytorch][bugfix]in variable_seq_lengths mode, set --log-throughput to false

ad32b3b5创建于 2025年10月10日历史提交

文件	最后提交记录	最后更新时间
chat_llama3_70b_ptd.sh	!1662 Llama2-mcore在线推理FA适配 Merge pull request !1662 from zhangjianxiang/master	1 年前
chat_llama3_8b_ptd.sh	!1662 Llama2-mcore在线推理FA适配 Merge pull request !1662 from zhangjianxiang/master	1 年前
ckpt_convert_llama3_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_llama3_mcore2hf.sh	!2828 [pytorch][sh]Add CUDA_DEVICE_MAX_CONNECTIONS Merge pull request !2828 from yanzhixiao/add-environment-variable	11 个月前
data_convert_llama3_instruction.sh	!2673 docs readme modify Merge pull request !2673 from jzh/docs_0517	1 年前
data_convert_llama3_instruction_pack.sh	!2673 docs readme modify Merge pull request !2673 from jzh/docs_0517	1 年前
data_convert_llama3_pairwise.sh	!1858 dpo、simpo方案特性支持：支持vpp、dpp、ep、cp、断点续训等 Merge pull request !1858 from glhyy/master	1 年前
data_convert_llama3_ppo.sh	!2054 【Ray PPO Part2】Support RewardWorker, RefWoker, ppo data preprocess Merge pull request !2054 from shishaoyu/master	1 年前
data_convert_llama3_pretrain.sh	!1656 llama3模型mcore适配 Merge pull request !1656 from yuhui/llama3	1 年前
data_convert_llama3_pretrain_pack.sh	!1806 Optim: llama3 qwen系列模型 预训练性能提升 Merge pull request !1806 from RuanZhiXiang/optimization-llama3-qwen	1 年前
dpo_llama3_8b_full_ptd.sh	!3163 [pytorch][feature]switch megatron_adaptor to v2 Merge pull request !3163 from yanzhixiao/switch-v2	9 个月前
dpo_llama3_8b_lora_ptd.sh	!3163 [pytorch][feature]switch megatron_adaptor to v2 Merge pull request !3163 from yanzhixiao/switch-v2	9 个月前
evaluate_llama3_70b_ptd.sh	!1662 Llama2-mcore在线推理FA适配 Merge pull request !1662 from zhangjianxiang/master	1 年前
evaluate_llama3_8b_ptd.sh	!1656 llama3模型mcore适配 Merge pull request !1656 from yuhui/llama3	1 年前
generate_llama3_70b_ptd.sh	!1662 Llama2-mcore在线推理FA适配 Merge pull request !1662 from zhangjianxiang/master	1 年前
generate_llama3_8b_lora_ptd.sh	[pytorch][bugfix]in variable_seq_lengths mode, set --log-throughput to false Co-authored-by: sunjunjie1587<sunjunjie8@huawei.com>	7 个月前
generate_llama3_8b_ptd.sh	!1662 Llama2-mcore在线推理FA适配 Merge pull request !1662 from zhangjianxiang/master	1 年前
pretrain_llama3_70b_ptd.sh	!1677 添加Qwen1.5-0.5B/1.8B迁移到mcore Merge pull request !1677 from caoruichao/master	1 年前
pretrain_llama3_70b_ptd_pack.sh	!2336 更新examples下的脚本 Merge pull request !2336 from xiecheng/master	1 年前
pretrain_llama3_8b_pack_ptd.sh	!2336 更新examples下的脚本 Merge pull request !2336 from xiecheng/master	1 年前
pretrain_llama3_8b_ptd.sh	!2336 更新examples下的脚本 Merge pull request !2336 from xiecheng/master	1 年前
tune_llama3_8b_full_pack.sh	!2306 修复新增patch导致patch顺序错乱报错的bug Merge pull request !2306 from shenjiarun/master	1 年前
tune_llama3_8b_full_ptd.sh	!2815 [pytorch][bugfix] fix logs bug Merge pull request !2815 from jzh/master_a	11 个月前
tune_llama3_8b_lora_ptd.sh	!2831 [pytorch][bugfix]update llama3-8b Merge pull request !2831 from jzh/master_b	11 个月前