Skip to content

Issues: pytorch/torchtitan

Beta
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

102 Open 235 Closed
102 Open 235 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
Assigned to nobody Loading
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Issues list

How to pretrain from scratch a Qwen 2.5 7B-base model using Torchtitan?
#1223 opened May 25, 2025 by tjoymeed
expert_bias is updated during training but saved checkpoint contains only zero values
#1222 opened May 24, 2025 by trestad
1
[Flux] Incorrect loss after loading from checkpoint
#1213 opened May 21, 2025 by CarlosGomes98
@wwwjn
6
[RFC] validation and evaluation in torchtitan
#1210 opened May 20, 2025 by tianyu-l
4
float8 rowwise vanilla TP low throughput bug Something isn't working module: float8
#1207 opened May 20, 2025 by danielvegamyhre
[MXFP8] unable to run titan llama3 debug model with mxfp8. Assertion: n_rows % max_row_tile_size == 0 bug Something isn't working
#1200 opened May 16, 2025 by lessw2020
@drisspg
4
Save RNG states during checkpointing for deterministic debugging enhancement New feature or request
#1194 opened May 14, 2025 by wwwjn
@wwwjn
document the usage of environment variables better_engineering Repo code quality improvements documentation Improvements or additions to documentation high priority triage review
#1192 opened May 14, 2025 by tianyu-l
PP Zero Bubble CI tests failure ci test failure high priority module: pipelining triage review
#1188 opened May 13, 2025 by tianyu-l
@H-Huang
2
issues on llama3 compile + (async) TP + AC ci test failure high priority module: torch.compile triage review
#1185 opened May 13, 2025 by tianyu-l
20
Can we support outputting checkpoints directly in .pt format? enhancement New feature or request module: checkpoint
#1177 opened May 9, 2025 by andrewor14
@fegin
8
how to inference with pretrained model?
#1169 opened May 6, 2025 by dragen1860
1
[Flux] Flux Issue Tracking
#1151 opened Apr 28, 2025 by wwwjn
4 of 16 tasks
@wwwjn
2
[Feature] Support validation
#1150 opened Apr 28, 2025 by CarlosGomes98
@wwwjn
4
[Question] FSDP+TP CUDA_DEVICE_MAX_CONNECTIONS documentation Improvements or additions to documentation module: fsdp question Further information is requested
#1147 opened Apr 27, 2025 by ChenchaoZhao
3
Inconsistent loss when resume training with vocab size that is not divisible by world size. high priority module: checkpoint triage review
#1136 opened Apr 23, 2025 by weixuansun
4
fully_shard() for huggingface model: pytorch caches too much GPU memory module: fsdp question Further information is requested
#1126 opened Apr 21, 2025 by mingdianliu
4
[DeepSeek MoE] current workstream planning enhancement New feature or request
#1125 opened Apr 21, 2025 by lessw2020
@lessw2020
1
Llama 4 issue tracking high priority triage review
#1118 opened Apr 17, 2025 by tianyu-l
3 of 13 tasks
6
Seeing - "Recomputed values for the following tensors have different metadata than during the forward pass." high priority triage review
#1117 opened Apr 17, 2025 by githubsgi
@soulitzer
19
FSDP2 root level parameter management module: fsdp question Further information is requested
#1091 opened Apr 11, 2025 by dingqingy
1
@tianyu-l
3
Torch.compile and TP during multiresolution Training module: torch.compile question Further information is requested
#1081 opened Apr 9, 2025 by nighting0le01
2
Is the currnet configuration system over-engineered? question Further information is requested
#1055 opened Apr 3, 2025 by wangkuiyi
2
Clarify PP split point documentation. question Further information is requested
#1054 opened Apr 3, 2025 by githubsgi
1
Overflow in F.scaled_dot_product_attention when using profiling with deterministic training
#1049 opened Apr 3, 2025 by JungHoyoun
1
ProTip! What’s not been updated in a month: updated:<2025-04-25.