Commits · b49168f883e0fb75cbb58e6f1031d1746da8bc98 · OpenDAS / AutoAWQ

20 Sep, 2023 30 commits
- Add bigcode fused modules (WIP) · b49168f8
  Casper Hansen authored Sep 20, 2023
  
  b49168f8
- Allow PytorchGELUTanh · dff3d157
  Casper authored Sep 21, 2023
  
  dff3d157
- Merge branch 'main' into bigcode · de8e7ffc
  Casper authored Sep 21, 2023
  
  de8e7ffc
- Merge pull request #62 from casper-hansen/refactor_quant · 72f954ce
  Casper authored Sep 21, 2023
```
Refactor quantization code
```
  72f954ce
- Disable fused layers · d6cf1442
  Casper Hansen authored Sep 20, 2023
  
  d6cf1442
- Remove cosine loss · 90bf52c7
  Casper Hansen authored Sep 20, 2023
  
  90bf52c7
- Correct comments · 9848b6a4
  Casper Hansen authored Sep 20, 2023
  
  9848b6a4
- Better comments. Implement cosine similarity · c9d01ac3
  Casper Hansen authored Sep 20, 2023
  
  c9d01ac3
- Move import up · e205548d
  Casper Hansen authored Sep 20, 2023
  
  e205548d
- Change filename · 2290bdb2
  Casper Hansen authored Sep 20, 2023
  
  2290bdb2
- Fix comment · 44b3187a
  Casper Hansen authored Sep 20, 2023
  
  44b3187a
- Fix small performance regression · 0f39807b
  Casper Hansen authored Sep 20, 2023
  
  0f39807b
- Simplify saving model · ed618bb0
  Casper Hansen authored Sep 20, 2023
  
  ed618bb0
- Retire entry.py and create examples/eval.py · 80996d1d
  Casper Hansen authored Sep 20, 2023
  
  80996d1d
- Remove old quantization code · d3550fec
  Casper Hansen authored Sep 20, 2023
  
  d3550fec
- Working for OPT · 724bda58
  Casper Hansen authored Sep 20, 2023
  
  724bda58
- Make code cleaner · 356cbc92
  Casper authored Sep 20, 2023
  
  356cbc92
- Add desc · 9204d6e8
  Casper authored Sep 20, 2023
  
  9204d6e8
- Added type hinting · 44470a7a
  Casper authored Sep 20, 2023
  
  44470a7a
- Initial quantization refactoring · 45c22ee5
  Casper authored Sep 20, 2023
  
  45c22ee5
- Remove comment · 19568c52
  Casper Hansen authored Sep 20, 2023
  
  19568c52
- Add GPT BigCode support (StarCoder) · 984fd2f8
  Casper Hansen authored Sep 20, 2023
  
  984fd2f8
- Merge pull request #60 from casper-hansen/kv_heads · a5e8b048
  Casper authored Sep 20, 2023
```
Support kv_heads
```
  a5e8b048
- Merge branch 'main' into kv_heads · a024e893
  Casper authored Sep 20, 2023
  
  a024e893
- Removed unused module · bf76e108
  Casper Hansen authored Sep 20, 2023
  
  bf76e108
- Use default attention shapes · d973a0e0
  Casper Hansen authored Sep 20, 2023
  
  d973a0e0
- Fix attention shapes. Add repeat interleave. · 2d593b84
  Casper Hansen authored Sep 20, 2023
  
  2d593b84
- Avoid fusing for larger Falcon models · e3936a44
  Casper Hansen authored Sep 20, 2023
  
  e3936a44
- Support kv_heads · 3256ffec
  Casper Hansen authored Sep 20, 2023
  
  3256ffec
- Merge pull request #58 from casper-hansen/gemmv2 · 63a12504
  Casper authored Sep 20, 2023
```
2x faster context processing with GEMV
```
  63a12504
19 Sep, 2023 3 commits
- Use GEMM v2 kernel for context processing · f3a71d1d
  Casper Hansen authored Sep 19, 2023
  
  f3a71d1d
- Move kernel into existing cuda file · 2fa3a5d1
  Casper Hansen authored Sep 19, 2023
  
  2fa3a5d1
- Add GEMM v2 kernel · 7cd461ff
  Casper Hansen authored Sep 19, 2023
  
  7cd461ff
15 Sep, 2023 7 commits
- Merge pull request #47 from casper-hansen/safetensors · 8788fe10
  Casper authored Sep 15, 2023
```
Safetensors and model sharding
```
  8788fe10
- Rename safetensors generation example · 4ecd8594
  Casper Hansen authored Sep 15, 2023
  
  4ecd8594
- Default to empty string for model file · 435b3b4b
  Casper Hansen authored Sep 15, 2023
  
  435b3b4b
- Merge pull request #27 from boehm-e/main · 2a3e0fa1
  Casper authored Sep 15, 2023
```
Allow user to use custom calibration data for quantization
```
  2a3e0fa1
- Add wikitext example · b9ab9a64
  Casper authored Sep 15, 2023
  
  b9ab9a64
- Move shuffling up for datasets loaded with load_dataset · d832a218
  Casper authored Sep 15, 2023
  
  d832a218
- Actually pass args to calib function · a9cef34b
  Casper authored Sep 15, 2023
  
  a9cef34b