Instructions to use shahils/GT-MHA with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use shahils/GT-MHA with Transformers:
# pip install -U transformers accelerate # Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("shahils/GT-MHA", device_map="auto") - Notebooks
- Google Colab
- Kaggle
Commit History
-
Rename TinyStories-v2 Checkpoints/MQA/checkpoint_step_249800.pt to TinyStories-v2 Checkpoints/MQA/checkpoint_step_250000.pt
Verified signature9bb7039
-
Correct TinyStories residual checkpoint to four-base run at step 250000
Verified signature27c0694
-
Add completed TinyStories QK-identity B4G8H16 seed0 checkpoint
Verified signature992b5ba -
Add completed TinyStories QK-identity B4G8H16 seed0 checkpoint
Verified signature9820fe4 -
Add completed TinyStories QK-identity B4G8H16 seed0 checkpoint
Verified signaturef59c1e2
-
Upload folder using huggingface_hub
Verified signature1054dc6
-
Update README.md
Verified signatureFixt GT-MHA model parameter size3fa5fa0
-
Remove redundant TinyStories checkpoint artifacts
Verified signaturea3d9cb4 -
Add LGMA package metadata
Verified signatureacd8902 -
Add TinyStories training entrypoint
Verified signature8b374a3 -
Add TinyStories sampling entrypoint
Verified signature243ce08 -
Add TinyStories checkpoint runtime
Verified signaturea0948b3 -
Add LGMA source for TinyStories checkpoint loading
Verified signaturec353cf5 -
Document safe TinyStories GT-MHA loading
Verified signature52aef04 -
Add TinyStories GT-MHA residual final report
Verified signature2bf3368 -
Add TinyStories GT-MHA residual final checkpoint
Verified signature1efe235
-
Update README.md
Verified signatureNo major changes0592f33 -
Add official GLUE results for GQA and collaborativeMHA
Verified signatureAdds official GLUE benchmark scores, parameter counts, and selection details for GQA and collaborativeMHA.c160bf6
-
Add files using upload-large-folder tool
Verified signature936b8a7 -
Add files using upload-large-folder tool
Verified signature9c353cc -
Add files using upload-large-folder tool
Verified signature97dc0bf -
Add files using upload-large-folder tool
Verified signature48e5f9a -
Add files using upload-large-folder tool
Verified signature3782828
-
Report official GLUE test results
Verified signaturec282628
-
Remove GLUE validation-loss table
Verified signaturef3cd930 -
Omit WNLI from GLUE validation tables
Verified signature9007729 -
Compile five-seed GLUE validation sweeps
Verified signaturea44ba0e
-
Update tuned GLUE sweep results
Verified signatureb8b0141
-
Update README with tuned five-seed STS-B results
Verified signatureae5dfb9 -
Add STS-B learning-rate selection record
Verified signature8cc22c1 -
Add STS-B learning-rate selection table
Verified signature39bef00 -
Add strict custom-attention checkpoint loader
Verified signatureca4b625 -
Add selected STS-B Collaborative MHA checkpoint
Verified signature25c006e -
Add selected STS-B GT-MHA residual checkpoint
Verified signature5a5736c -
Add selected STS-B GQA checkpoint
Verified signature31fea48 -
Add selected STS-B MHA checkpoint
Verified signaturefafcb90 -
Report complete three-seed GLUE suite
Verified signaturea8ea404 -
Add four-task GLUE macro average
Verified signatureb7f00d3 -
Report best validation within four epochs
Verified signature4d35674 -
Restore three-seed mean and std GLUE results
Verified signature36a1082 -
Report best GLUE seed for every attention method
Verified signature3c317f2 -
Report best GT-MHA residual GLUE seeds
Verified signature31bd8dc -
Report GT-MHA residual three-seed GLUE results
Verified signature2a25441 -
Report GLUE scores on 100-point scale
Verified signatureef1574f -
Report Collaborative MHA three-seed GLUE results
Verified signature8af5c46
-
Report GQA three-seed GLUE results
Verified signatureadb0ec3 -
Report MHA three-seed GLUE mean and std
Verified signature4af9eb5 -
Add corrected Collaborative MHA and GT-MHA residual GLUE results
Verified signature32042df -
Add corrected Collaborative MHA cola final checkpoint
Verified signature715750a -
Add corrected Collaborative MHA sst2 final checkpoint
Verified signaturedab022b