refiners
refiners copied to clipboard
finegrain-ai
A microframework on top of PyTorch with first-class citizen APIs for foundation model adaptation
This is an early draft PR to indicate that i'm working on the fuyu 8b implementation. A lot of work still needs to be done, including - [x] Finalization of...
Currently working on Mistral-7b 0.1 model: - [X] Implement Mistral-7B 0.1 - [X] Conversion - [X] Unit Test
This makes it so we can do lower precision training. @piercus might be interested
This piece of code helped me identify how much time is spent in the data processing stage, the forward pass, and backward pass so that I know where to optimize....
## Context Still working on #165 Want to explore the lora/non-lora in the cross-attention adapter and to replace, conditionnaly ```python fl.Linear( ... bias=text_cross_attention.use_bias ) ``` with ```python fl.LinearLora( ... bias=text_cross_attention.use_bias...
Adds a seamless tile mode but does so in a way you guys probably won't like :-) Opening the pull request to have the discussion. Tile mode can be set...
## Context Made on top of #179 for the dataset Related to #165 Design doc in https://docs.google.com/document/d/1JD-85slm_hJCIBCAGXyTVxkbPiPmS2N0ZzSy_iiube0/edit#heading=h.3lay6geakhbf ## Test Status This PR is running a 2x8GB GPU lora training ```bash...
Still working on the LLaVa architecture. It tooks longer than expected to understand the architecture and how the generate method works on a multimodal aspect. I'll go with the huggingface...
This is my draft PR for the bounty **ELLA for SD1.5**. I implemented and tested the architecture. I have few steps left. - [x] Implement ELLA - [x] Test the...
Seems to work quite well sometimes for removing text, logos, and titles on photos on the covers of magazines. would be very useful
Metadata
Owner
Metadata
A microframework on top of PyTorch with first-class citizen APIs for foundation model adaptation