k2SSL: A Faster and Better Framework for Self-Supervised Speech Representation Learning - #1745
k2SSL: A Faster and Better Framework for Self-Supervised Speech Representation Learning#1745yfyeung wants to merge 56 commits into
Conversation
update Update ssl_datamodule.py Update pretrain.py Update pretrain.sh Update pretrain.sh Update hubert_ce.py Update pretrain.py
|
Part of the resources have been released: Zipformer Base pre-trained with cross entropy loss: Checkpoints, logs, and scripts With these resources, I believe anyone with 8 V100 32G GPUs can easily reproduce our experiments. |
35fd0cf to
1b89c6d
Compare
|
Important Review skippedDraft detected. Please check the settings in the CodeRabbit UI or the You can disable this status message by setting the 🪧 TipsChatThere are 3 ways to chat with CodeRabbit:
SupportNeed help? Create a ticket on our support page for assistance with any issues or questions. Note: Be mindful of the bot's finite context window. It's strongly recommended to break down tasks such as reading entire modules into smaller chunks. For a focused discussion, use review comments to chat about specific files and their changes, instead of using the PR comments. CodeRabbit Commands (Invoked using PR comments)
Other keywords and placeholders
CodeRabbit Configuration File (
|
|
Thanks for the great work! Do we consider merging the PR and so icefall offically support it? Thanks |
Paper link: https://arxiv.org/abs/2411.17100v1