Commits · a2aa804ccdcd52e6aec09fac07727061cda66ad7 · OpenDAS / AutoAWQ

12 Sep, 2023 8 commits
- Falcon 40B attention shapes [WIP] · 28d52d81
  Casper Hansen authored Sep 12, 2023
  
  28d52d81
- xq view key · bc4f93c2
  Casper Hansen authored Sep 12, 2023
  
  bc4f93c2
- Fix Falcon inputs ids · ba4da393
  Casper Hansen authored Sep 12, 2023
  
  ba4da393
- Falcon: Enable fused attention · f03f72de
  Casper Hansen authored Sep 12, 2023
  
  f03f72de
- Create custom attention shape for Falcon 7B · ac770875
  Casper Hansen authored Sep 12, 2023
  
  ac770875
- xk_reshape key · 7f8f9f16
  Casper Hansen authored Sep 12, 2023
  
  7f8f9f16
- Refactor view/reshaping into a predefined dict · a8c9afd5
  Casper Hansen authored Sep 12, 2023
  
  a8c9afd5
- Create Falcon block and model for fusing · 4517b3f2
  Casper Hansen authored Sep 12, 2023
  
  4517b3f2
11 Sep, 2023 8 commits
- Remove xformers · 1aa8aebd
  Casper Hansen authored Sep 11, 2023
  
  1aa8aebd
- Update MPTBlock, fuse with MPTModel · dd41a223
  Casper Hansen authored Sep 11, 2023
  
  dd41a223
- Fix attention, support alibi · 7631add1
  Casper Hansen authored Sep 11, 2023
  
  7631add1
- Create MPTModel class · 90f54dbb
  Casper Hansen authored Sep 11, 2023
  
  90f54dbb
- Initialize with device · e80663bb
  Casper Hansen authored Sep 11, 2023
  
  e80663bb
- Remove fusing attention, only blocks · ac3e86df
  Casper Hansen authored Sep 11, 2023
  
  ac3e86df
- Fuse MPT block · 950851b3
  Casper Hansen authored Sep 11, 2023
  
  950851b3
- Switch to torch SDPA · d7badefc
  Casper Hansen authored Sep 11, 2023
  
  d7badefc
09 Sep, 2023 7 commits
- Falcon fused layers · 54f02854
  Casper Hansen authored Sep 09, 2023
  
  54f02854
- Fix alibi, rotary arg, neox, arg · 73c5e2bf
  Casper Hansen authored Sep 09, 2023
  
  73c5e2bf
- Fuse MPT · f3695d60
  Casper Hansen authored Sep 09, 2023
  
  f3695d60
- Update module name · 5bd6fbc7
  Casper Hansen authored Sep 09, 2023
  
  5bd6fbc7
- Implement ALiBi. · adc5304b
  Casper Hansen authored Sep 09, 2023
  
  adc5304b
- Remove warnings about GEMM · c58ec73b
  Casper Hansen authored Sep 09, 2023
  
  c58ec73b
- Create torch SDPA implementation · ebbbc3a5
  Casper Hansen authored Sep 09, 2023
  
  ebbbc3a5
08 Sep, 2023 10 commits
- GEMM + GEMV compatibility · 890b6aa7
  Casper Hansen authored Sep 08, 2023
  
  890b6aa7
- Use apply_rotary_emb · 64e6b3e1
  Casper Hansen authored Sep 08, 2023
  
  64e6b3e1
- Update kernel, remove unused code · a11c313a
  Casper Hansen authored Sep 08, 2023
  
  a11c313a
- Attention works · fbfa9d82
  Casper Hansen authored Sep 08, 2023
  
  fbfa9d82
- Fix MLP module · 75710806
  Casper Hansen authored Sep 08, 2023
  
  75710806
- Fix variables and no contiguous memory for GEMV · e197a733
  Casper Hansen authored Sep 08, 2023
  
  e197a733
- Fix variables · 6534f5e6
  Casper Hansen authored Sep 08, 2023
  
  6534f5e6
- Implement GEMM/GEMV in quantize function and fused modules · 5db43a7f
  Casper Hansen authored Sep 08, 2023
  
  5db43a7f
- Add deprecation warning · 9b2946b6
  Casper Hansen authored Sep 08, 2023
  
  9b2946b6
- Refactor modules, create separate GEMM and GEMV · fe314160
  Casper Hansen authored Sep 08, 2023
  
  fe314160
07 Sep, 2023 1 commit
- Support Falcon 180B · 475481b4
  Casper authored Sep 07, 2023
  
  475481b4
06 Sep, 2023 6 commits
- Bump version · a0fb91d9
  Casper authored Sep 06, 2023
  
  a0fb91d9
- Remove simple dispatch · 562c0d52
  Casper authored Sep 06, 2023
  
  562c0d52
- Multi-GPU support for quantized models · f8273a0c
  Casper Hansen authored Sep 06, 2023
  
  f8273a0c
- Get correct devices · 7cf0d987
  Casper Hansen authored Sep 06, 2023
  
  7cf0d987
- Correct input sizes · ab7d68e7
  Casper Hansen authored Sep 06, 2023
  
  ab7d68e7
- Switch to model.generate() · a4626828
  Casper Hansen authored Sep 06, 2023
  
  a4626828