Skip to content

Joint text and audio dataloading for speechllm#25

Open
pzelasko wants to merge 17 commits intozhehuaichen:canary_speechllm1_cross_t5_pr3from
pzelasko:canary_speechllm1_cross_t5_pr3
Open

Joint text and audio dataloading for speechllm#25
pzelasko wants to merge 17 commits intozhehuaichen:canary_speechllm1_cross_t5_pr3from
pzelasko:canary_speechllm1_cross_t5_pr3

Conversation

@pzelasko
Copy link

@pzelasko pzelasko commented May 8, 2024

What does this PR do ?

Add a one line overview of what this PR aims to accomplish.

Collection: [Note which collection this PR will affect]

Changelog

  • Add specific line by line info of high level changes in this PR.

Usage

  • You can potentially add a usage example below
# Add a code snippet demonstrating how to use this 

Before your PR is "Ready for review"

Pre checks:

  • Make sure you read and followed Contributor guidelines
  • Did you write any new necessary tests?
  • Did you add or update any necessary documentation?
  • Does the PR affect components that are optional to install? (Ex: Numba, Pynini, Apex etc)
    • Reviewer: Does the PR have correct import guards for all optional libraries?

PR Type:

  • New Feature
  • Bugfix
  • Documentation

If you haven't finished some of the above items you can still open "Draft" PR.

Who can review?

Anyone in the community is free to review the PR once the checks have passed.
Contributor guidelines contains specific people who can review PRs to various areas.

Additional Information

  • Related to # (issue)

pzelasko added 3 commits May 8, 2024 10:25
Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com>
Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com>
Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com>
zhehuaichen and others added 14 commits May 8, 2024 19:47
predict

Signed-off-by: zhehuaichen <dian.chenzhehuai@gmail.com>
…lhotse

include text-only into fprop of training and eval; TODO: text-only inference
Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com>
Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com>
Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com>
Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com>
…omized_round_robin

Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com>
…own batch settings that can be merged with zip sampler to enjoy max batch sizes for both modalities in a single training step. Each modality runs fwd+bwd in turn to save GPU memory (instead of running fwd separately and bwd together).

Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com>
Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com>
Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com>
Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com>
Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com>
Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants

Comments