andreaskoepf
Follow
Highlights
- Pro
Popular repositories
-
faster-rcnn.torch Public
Experimental Torch7 implementation of RCNN for Object Detection with a Region Proposal Network
-
mlp-hidden-visu Public
Visualization of hidden layer activations of small multilayer perceptrons (MLPs)
-
-
-
-
CLIP-Image-Captioning Public
Forked from TheoCoombes/ClipCap
[WIP] Using CLIP and GPT2-XL to generate captions from images.
1,105 contributions in the last year
Contribution activity
April 2023
Created 14 commits in 2 repositories
Created a pull request in LAION-AI/Open-Assistant that received 12 comments
Improve scores of small 1.4B reward model..
add shf & hellaswag loading options
add WebGPTRank
correct default oasst-rm-1-pythia-1.4b configuration
remove RM support of HFDataset in favor of H…
+249
−160
•
12
comments
Opened 9 other pull requests in 1 repository
LAION-AI/Open-Assistant
5
merged
3
open
1
closed
- Add option to export trlx checkpoints
- Model training dockerfile
- Add HFSummaryPairs class & fix AnthropicRLHF parsing
- Use fixed RNG seed value for all DeepSpeed workers
- Code state of wandb run bqiatai0 (2023-03-25_oasst_cyrillic_alpaca_reference_good)
- ignore, see #2312 for correct version
- Revert unrelated changes in instructor rank_datasets
- Prepare oasst data export to HuggingFace
- Fix loading of Nebulous/gpt4all_pruned dataset
Reviewed 24 pull requests in 1 repository
LAION-AI/Open-Assistant
24 pull requests
- Add more German translations
- Reduced max tokens of llama 30b to 1792 because of OOMs at 2048
- Add alpaca reverse augmentation possibility
- Add vicuna dataset
- fix types for prompt dataset
- update gpt4all to add multiround
- Add trailing slashes to backend calls
- hotfix for prod deployment
- Updated nginx config for prod, including streaming headers
- fix typos
- Improve scores of small 1.4B reward model..
- fix deepspeed issue on trainer_rm.py, add crossentropy support
- Disable Initial Prompt Task for en and es Locales
- Update configs according to feedback to #2277
- fix list index out of range error for soda dataset
- inference: use uuid v7 for most of table
- Flash attention support for Llama
- Fix DeepSpeed 0.8.3 training
- add dataset counts script
- Reward Model evaluation
- fix for hf dockerfile
- revert instructor code and fix a bug in anthropic ds parsing
- Set validation max length to a different value.
- two more datasets
Created an issue in LAION-AI/Open-Assistant that received 1 comment
Integrate reward/instructor implementation into trainer_rm
Currently we have two different reward model training code branches, e.g. model/model_training/trainer_rm.py vs. model/reward/instructor/trainer.py…
1
comment