Ecosyste.ms: Issues
An open API service for providing issue and pull request metadata for open source projects.
GitHub / iejmac/encoder-distill issues and pull requests
#19 - Alignment might be easier with word sense disambiguated and/or averaged vectors
Issue -
State: open - Opened by Thomas-MMJ over 2 years ago
#18 - Similarity Loss: Softer guidance via similarities instead of features
Pull Request -
State: closed - Opened by iejMac over 2 years ago
#17 - Research Priority Queue
Issue -
State: open - Opened by iejMac over 2 years ago
Labels: research
#16 - Hard guidance lower in the network and soft guidance higher
Issue -
State: open - Opened by iejMac over 2 years ago
Labels: research
#15 - Add MLP to teacher instead of student
Issue -
State: open - Opened by iejMac over 2 years ago
- 1 comment
Labels: research
#15 - Add MLP to teacher instead of student
Issue -
State: open - Opened by iejMac over 2 years ago
- 1 comment
Labels: research
#14 - Try to use similarity as a loss
Issue -
State: open - Opened by rom1504 over 2 years ago
- 3 comments
Labels: research
#13 - Try to do distill for more GPU hours
Issue -
State: open - Opened by rom1504 over 2 years ago
Labels: research
#12 - Try to ensemble 2 clip by targeting their sum
Issue -
State: open - Opened by rom1504 over 2 years ago
Labels: research
#11 - Normalize features before MSE Loss
Pull Request -
State: closed - Opened by iejMac over 2 years ago
- 1 comment
#11 - Normalize features before MSE Loss
Pull Request -
State: closed - Opened by iejMac over 2 years ago
- 1 comment
#10 - generalize MLPEncoder stuff
Issue -
State: open - Opened by iejMac over 2 years ago
#9 - combine separately trained encoders into CLIP model
Pull Request -
State: closed - Opened by iejMac over 2 years ago
#8 - Batch size scaling
Issue -
State: open - Opened by iejMac over 2 years ago
- 3 comments
Labels: research
#7 - try normalizing before MSE
Issue -
State: open - Opened by iejMac over 2 years ago
Labels: research
#6 - control steps and warmup using params
Pull Request -
State: closed - Opened by iejMac over 2 years ago
#5 - write function that combines image and text back into CLIP
Issue -
State: closed - Opened by iejMac over 2 years ago
#4 - organize scripts + split training for each encoder
Pull Request -
State: closed - Opened by iejMac over 2 years ago
- 1 comment
#3 - clean and package up
Issue -
State: open - Opened by iejMac over 2 years ago
#2 - align internal layers of model as well as final layer
Issue -
State: open - Opened by iejMac over 2 years ago
Labels: research
#1 - make script train encoders separately
Issue -
State: closed - Opened by iejMac over 2 years ago