Ecosyste.ms: Issues

An open API service for providing issue and pull request metadata for open source projects.

GitHub / shan18/perceiver-resampler-xattn-captioning issues and pull requests

#41 - Add MLP module in Mapping Network

Pull Request - State: closed - Opened by rvk007 about 2 years ago

#40 - Tokenizer

Pull Request - State: closed - Opened by Charithaakula about 2 years ago

#39 - add support for multiple sign language video loading in the dataset

Pull Request - State: closed - Opened by shan18 about 2 years ago

#38 - Add support for different decoding strategies in inference

Pull Request - State: closed - Opened by shan18 about 2 years ago

#37 - Add mapping layer to map clip embeddings to gpt inputs

Pull Request - State: closed - Opened by Charithaakula about 2 years ago

#36 - Add options to disable resampler and enable lm head trainable state

Pull Request - State: closed - Opened by shan18 about 2 years ago

#35 - Create a mapping network to change CLIP embeddings in GPTs distribution

Issue - State: closed - Opened by shan18 about 2 years ago - 1 comment
Labels: enhancement, high priority

#34 - Update dataloader to output multilingual sign language videos

Issue - State: closed - Opened by shan18 about 2 years ago
Labels: enhancement

#33 - Sending tokens in masked fashion to GPT2

Pull Request - State: closed - Opened by shan18 about 2 years ago

#32 - Added support for gated cross attention layers in GPT2

Pull Request - State: closed - Opened by rvk007 about 2 years ago

#31 - Look for attention masking for padded frames in resampler

Issue - State: closed - Opened by shan18 about 2 years ago
Labels: high priority

#30 - Support for saving and restoring model state

Pull Request - State: closed - Opened by shan18 about 2 years ago

#29 - Remove pad tokens from BLEU calculation

Issue - State: closed - Opened by shan18 about 2 years ago - 1 comment
Labels: bug

#28 - Create script for model checkpoint averaging

Issue - State: closed - Opened by shan18 about 2 years ago
Labels: enhancement, wontfix

#27 - Ignore pad token id in loss function

Pull Request - State: closed - Opened by shan18 about 2 years ago

#26 - BLEU-4 Integration

Pull Request - State: closed - Opened by shan18 about 2 years ago

#25 - Custom video embed

Pull Request - State: closed - Opened by Charithaakula about 2 years ago

#24 - Ignore pad idx from cross entropy loss function

Issue - State: closed - Opened by shan18 about 2 years ago

#23 - Add support for multilingual text generation

Issue - State: closed - Opened by shan18 about 2 years ago
Labels: not immediate, enhancement, wontfix

#22 - Add option to resume training from a previous checkpoint

Issue - State: closed - Opened by shan18 about 2 years ago
Labels: not immediate

#21 - Inspect usage of ff_mult in resampler

Issue - State: closed - Opened by rvk007 about 2 years ago - 2 comments
Labels: information

#20 - Integrating hydra to the codebase

Pull Request - State: closed - Opened by shan18 about 2 years ago

#19 - add docs to resampler code

Pull Request - State: closed - Opened by rvk007 about 2 years ago

#18 - Videospecific pos emb

Pull Request - State: closed - Opened by Charithaakula about 2 years ago

#17 - Revert "Adding videospecific position embedding"

Pull Request - State: closed - Opened by Charithaakula about 2 years ago

#16 - Adding videospecific position embedding

Pull Request - State: closed - Opened by Charithaakula about 2 years ago

#15 - Adding support for custom runs - First cut

Pull Request - State: closed - Opened by Charithaakula about 2 years ago

#14 - Add support for training with 8 num_workers

Issue - State: closed - Opened by shan18 about 2 years ago - 1 comment
Labels: not immediate, wontfix

#13 - Optimizing the number of parameters in the model

Issue - State: closed - Opened by shan18 about 2 years ago - 1 comment

#12 - Hydra integration

Issue - State: closed - Opened by shan18 about 2 years ago - 1 comment

#11 - Add support for wandb logging

Issue - State: closed - Opened by rvk007 about 2 years ago
Labels: enhancement, wontfix

#10 - Add inference mode to the trainer class

Issue - State: closed - Opened by shan18 about 2 years ago
Labels: high priority

#9 - Add training engine

Pull Request - State: closed - Opened by rvk007 about 2 years ago

#8 - Add BLEU score metric

Issue - State: closed - Opened by rvk007 about 2 years ago

#7 - Docstrings for Perceiver Resampler

Issue - State: closed - Opened by shan18 about 2 years ago

#6 - Add Perceiver Resampler module

Pull Request - State: closed - Opened by rvk007 about 2 years ago

#4 - Add the checkpointing functionality to the trainer

Issue - State: closed - Opened by shan18 about 2 years ago

#3 - Build the text generator

Issue - State: closed - Opened by shan18 about 2 years ago - 1 comment

#2 - Build the trainer function

Issue - State: closed - Opened by shan18 about 2 years ago

#1 - Send max timesteps length for positional encoding from the dataset

Issue - State: closed - Opened by shan18 about 2 years ago - 1 comment