Skip to content

Issues: alibaba/Pai-Megatron-Patch

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

36 Open 132 Closed
36 Open 132 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Assignee
Filter by who’s assigned
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Issues list

Pretrain data processing will drop samples that containing special tokens?
#403 opened Dec 22, 2024 by juncaofish
tensor model parallel isn't handled properly for shared_expert_gate in Qwen2-MoE and kv_a_proj_with_mqa in DeepSeekV2
#397 opened Dec 15, 2024 by cosmosZhou
2 tasks
1
2
Qwen2.5-7B SFT NaN in local forward loss calculation
#392 opened Dec 5, 2024 by Eisenhower
1
qwen2.5 1.5B 无法正常训练
#386 opened Nov 24, 2024 by Traveller2001
2
PAI平台上训练llama-3.1-8b时有一个GPU使用率偏低
#383 opened Nov 20, 2024 by kkkeepgoing
1
RuntimeError: DataLoader worker exited unexpectedly
#378 opened Nov 15, 2024 by kkkeepgoing
1
关于qwen1.5-moe的数据格式问题
#375 opened Nov 4, 2024 by cdxzyc
最新版本是否支持conversation-sft，多轮对话？
#369 opened Oct 24, 2024 by gaohejiao
DeepSeekV2Tokenizer should use padding_side="right" in __init__()!
#368 opened Oct 23, 2024 by pqhgit
4
AssertionError: Rank 11: found NaN in local grad norm in backward pass before data-parallel communication collective. Device: 3
#366 opened Oct 21, 2024 by lanfengmo
3
关于LLAMA 3.1模型的适配问题
#361 opened Oct 14, 2024 by echo-valor
1
对qwen-2.5扩充词表后loss飙升
#360 opened Oct 9, 2024 by QianguoS
1
cannot import name 'TEDotProductAttentionMLA' when running examples/deepseek_v2/run_mcore_deepseek.sh
#359 opened Oct 4, 2024 by dreasysnail
4
llama3.1-8b转换megatron-mcore格式后模型大小从15G变成了71G，精度仍然为bf16，这是正常的吗
#356 opened Sep 26, 2024 by kkkeepgoing
3
qwen2.5转换脚本转换时报错
#354 opened Sep 25, 2024 by enze5088
1
加入群聊失败， 第二个群也不能扫码加入了
#351 opened Sep 23, 2024 by GeorgeSen
7
llama3.1 8b训练32k的上下文模型，训练时间长、并且loss偏大
#348 opened Sep 20, 2024 by ARQlalala
1
有适配minicpm的打算吗？
#345 opened Sep 13, 2024 by adol001
建议对deepseek-v2-coder-lite进行sft测试
#342 opened Sep 12, 2024 by bao-xiaoyi
5
DeepSeek Vocab-size Mismatch
#338 opened Sep 10, 2024 by Jiayi-Pan
1
1
AssertionError: First dimension of the tensor should be divisible by tensor parallel size
#332 opened Sep 4, 2024 by pizts
deepseek模型转换问题
#327 opened Aug 25, 2024 by bao-xiaoyi
7
qwen2-sft 训练起步阶段就卡住
#325 opened Aug 22, 2024 by baisechundu
7
Channel Loss支持
#316 opened Aug 7, 2024 by echo-valor
2
optimizer offloading 太强了
#311 opened Aug 2, 2024 by 154912369
2
ProTip! Add no:assignee to see everything that’s not assigned.