Modalities modalities issues

Modalities / modalities

A framework for training multimodal foundation models.

MIT License

38 stars 2 forks source link

issues

Newest

Newest Most commented Recently updated Oldest Least commented Least recently updated

Configurable Training Length

#175 flxst opened 1 day ago
0
Support and test python versions 3.10 and 3.11

#174 flxst opened 2 days ago
0
Supported Python Versions

#173 flxst opened 2 days ago
0
Pull Request & Issue Templates

#172 flxst opened 2 days ago
0
Fix getting started example

#171 flxst closed 2 days ago
0
SwiGLU naming of projection matrices

#170 le1nux opened 4 days ago
0
feat: log config

#169 mali-git opened 4 days ago
0
Draft: Feat/initialization component

#168 le1nux closed 3 days ago
0
Improve the dataset inheritance and class naming

#167 le1nux opened 1 week ago
1
Fix getting started example

#166 flxst closed 1 week ago
0
Limited and potentially incorrect weight initialization for CoCa model

#165 flxst closed 3 days ago
1
Fix/dataset index: Index values were faulty when indexing the original samples instead of blocks.

#164 le1nux closed 6 days ago
2
Bug: Dataset implementation does not

#163 le1nux opened 1 week ago
0
Disable Flash Attention for inference

#162 rrutmann opened 1 week ago
1
Feat: Various Configurable Initializations

#161 flxst opened 1 week ago
0
Generic downstream eval

#160 mrudat-iais opened 1 week ago
0
Make Activation Checkpointing Configurable

#159 mali-git closed 1 week ago
2
Fix/sequence length power of 2

#158 le1nux closed 6 days ago
0
Fix CPU Tests

#157 flxst closed 2 weeks ago
0
Revision of block size and sequence length

#156 flxst opened 2 weeks ago
0
Unified unit & end-to-end testing

#155 flxst closed 2 weeks ago
2
Manual SwiGLU implementation

#154 mali-git closed 2 weeks ago
1
Various small fixes for Mamba, unit tests & linting

#153 flxst closed 3 weeks ago
0
Tokenizer remove max length flag

#152 le1nux closed 3 weeks ago
1
feat: implemented scaled weight initialization

#151 le1nux closed 3 weeks ago
0
New downstream eval

#150 mrudat-iais closed 1 week ago
1
Add Missing Dependency

#149 mali-git closed 3 weeks ago
0
mamba_ssm dependency missing

#148 le1nux opened 3 weeks ago
0
Scaled weight init

#147 le1nux closed 3 weeks ago
0
Add topk sampling for generation

#146 fromm-m opened 3 weeks ago
0
Dev experiments max

#145 le1nux closed 3 weeks ago
0
Feat/multiple attention implementations into dev_experiments_max

#144 le1nux closed 4 weeks ago
0
Feat/optimizer parameter groups into dev_experiments_max

#143 le1nux closed 4 weeks ago
0
Optimizer parameter groups

#142 le1nux closed 4 weeks ago
0
Towards stable modalities version

#141 le1nux opened 1 month ago
0
fix: Support different versions of accelerate

#140 rrutmann closed 1 month ago
0
Implementation of optimizer parameter groups with and without weight decay

#139 flxst closed 4 weeks ago
0
Implementation of multiple attention mechanisms

#138 flxst closed 3 weeks ago
0
Fix Training Step Logging & Log Number of Consumed Tokens

#137 mali-git closed 1 month ago
3
Fix Pre-LN for GPT2 Model

#136 flxst closed 1 month ago
0
Include torch compile

#135 flxst opened 1 month ago
0
fix: #97 Unused attributes in a component's config yaml are just ignored by Pydantic

#134 lllAlexanderlll closed 3 weeks ago
2
Apply Torch-Compile

#133 mali-git opened 1 month ago
0
Refactor RMSNorm Implementation

#132 mali-git opened 1 month ago
1
fix: deactivation of model evaluation mode

#131 flxst closed 1 month ago
0
First optimizer changes towards improved loss stability

#130 le1nux closed 2 weeks ago
0
Investigation: Loss spikes when loss developes close to convergence

#129 le1nux opened 1 month ago
1
Resolve Model Class Within Modalities

#128 mali-git closed 3 weeks ago
0
debug: adapt to new accelerate API

#127 mali-git closed 1 month ago
0
Fix Bug when using RoPE embeddings together with gradient activation

#126 le1nux opened 1 month ago
1