binmakeswell

icon indicating an email [email protected]

icon company @hpcaitech icon location Singapore icon location Stay Hungry, Stay Foolish.

Results 359 comments of binmakeswell

[BUG]: ZeRO causes runtime error when use GRU and pack sequence

We have updated a lot. This issue was closed due to inactivity. Thanks.

[BUG]: Acc decreases when training with PP and TP

We have updated a lot. This issue was closed due to inactivity. Thanks.

ZeRO dose not initialize weight correctly

We have updated a lot. This issue was closed due to inactivity. Thanks.

[BUG]: Relationship between `BATCH_SIZE`, `LEARNING_RATE` and GPU amount

This issue was closed due to inactivity. Thanks.

[FEATURE]: example - Diffusion Model

Done. https://github.com/hpcaitech/ColossalAI/tree/main/examples/images/diffusion

Questions about log interpretation, seems paradoxical

We have updated a lot. This issue was closed due to inactivity. Thanks.

[BUG]: found inf during ShardedOptimV2 step

We have updated a lot. This issue was closed due to inactivity. Thanks.

[BUG]: ZeRO not Working with SGD Optimizer

We have updated a lot. This issue was closed due to inactivity. Thanks.

[DOC]: Need a simple Colab sample to show CUDA and Torch versions

We have updated a lot. Colab is not suitable for large-scale parallelism. This issue was closed due to inactivity. Thanks.

[FEATURE]: can Colossal-AI support customized parallel schedule? is there a tutorial for that?

We have updated a lot. This issue was closed due to inactivity. Thanks.