Commits · 29ee66d9e77f3e443d48a17b4838d00a76bc6f5e · OpenDAS / AutoAWQ

03 Feb, 2024 3 commits
- PEFT compatible GEMM (#324) · 29ee66d9
  Casper authored Feb 03, 2024
  
  29ee66d9
- Marlin symmetric quantization and inference (#320) · 34085edc
  Ilyas Moutawwakil authored Feb 03, 2024
```
Co-authored-by: Casper <casperbh.96@gmail.com>
```
  34085edc
- AMD ROCM Support (#315) · f018d2b7
  Ilyas Moutawwakil authored Feb 03, 2024
  
  f018d2b7
26 Jan, 2024 1 commit
- Fixes a breaking change in autoawq (#325) · 8c78db4d
  Younes Belkada authored Jan 26, 2024
  
  8c78db4d
24 Jan, 2024 3 commits
- Evaluation: Add more evals (#283) · 7133575a
  Casper authored Jan 24, 2024
  
  7133575a
- Up to 60% faster context processing (#316) · 149236e4
  Casper authored Jan 24, 2024
  
  149236e4
- Torch only inference + any-device quantization (#319) · c6c7b065
  Casper authored Jan 24, 2024
  
  c6c7b065
21 Jan, 2024 1 commit
- Exllama kernels support (#313) · 2fcbf268
  Ilyas Moutawwakil authored Jan 21, 2024
```
Co-authored-by: Casper <casperbh.96@gmail.com>
```
  2fcbf268
28 Dec, 2023 1 commit
- AWQ: Separate the AWQ kernels to separate repository (#279) · 2edb3f6f
  Casper authored Dec 28, 2023
```
Co-authored-by: Casper Hansen <casperbh96@gmail.com>
```
  2edb3f6f
23 Dec, 2023 2 commits
- Set default rope_theta on LlamaLikeBlock (#271) · 66adba4a
  Casper authored Dec 23, 2023
  
  66adba4a
- Add Baichuan2 Support (#247) · cef9f113
  Aoyu authored Dec 23, 2023
```
Co-authored-by: Casper <casperbh.96@gmail.com>
```
  cef9f113
22 Dec, 2023 2 commits
- Fused rope theta (#270) · 727172e9
  Casper authored Dec 23, 2023
```
Co-authored-by: Casper Hansen <casperbh96@gmail.com>
```
  727172e9
- Mixtral: Mixture of Experts quantization (#251) · 5b9f3c47
  Casper authored Dec 22, 2023
  
  5b9f3c47
11 Dec, 2023 1 commit
- [`core`] Make AutoAWQ fused modules compatible with HF transformers (#244) · ae24f424
  Younes Belkada authored Dec 11, 2023
  
  ae24f424
15 Nov, 2023 2 commits
- [`core`] Add `is_hf_transformers` flag (#195) · 74d0fe44
  Younes Belkada authored Nov 15, 2023
  
  74d0fe44
- [`core`] Replace `QuantLlamaMLP` with `QuantFusedMLP` (#188) · 3b362c0d
  Younes Belkada authored Nov 15, 2023
  
  3b362c0d
14 Nov, 2023 1 commit
- Fix multi-GPU loading and inference (#190) · 09c73fb2
  Casper authored Nov 14, 2023
  
  09c73fb2
11 Nov, 2023 2 commits
- Adaptive batch sizing (#181) · c5581b27
  Casper authored Nov 11, 2023
  
  c5581b27
- Reset cache on new generation (#178) · df909e83
  Casper authored Nov 11, 2023
  
  df909e83
10 Nov, 2023 1 commit
- New logic for passing past_key_value (#177) · 5db86ec5
  Younes Belkada authored Nov 10, 2023
  
  5db86ec5
04 Nov, 2023 1 commit
- Create fused LlamaLikeModel (#152) · 8110e028
  Casper authored Nov 04, 2023
  
  8110e028
03 Nov, 2023 3 commits
- Fix condition when rolling cache (#150) · 958678df
  Casper authored Nov 04, 2023
  
  958678df
- [`core` / `attention`] Fix fused attention generation with newest transformers version (#146) · 92a403b2
  Younes Belkada authored Nov 04, 2023
```
Co-authored-by: Casper <casperbh.96@gmail.com>
```
  92a403b2
- Fix performance regression (#148) · 3eda6562
  Casper authored Nov 03, 2023
  
  3eda6562
28 Oct, 2023 1 commit
- [`core`] Support fp32 / bf16 inference (#121) · 18712d00
  Younes Belkada authored Oct 28, 2023
  
  18712d00
07 Oct, 2023 1 commit
- Only apply attention mask if seqlen is greater than 1 · e94b7f40
  Casper Hansen authored Oct 07, 2023
  
  e94b7f40
06 Oct, 2023 7 commits
- Remove attention sinks · b13e2a85
  Casper Hansen authored Oct 06, 2023
  
  b13e2a85
- Create RoPE module · 69733d2c
  Casper Hansen authored Oct 06, 2023
  
  69733d2c
- Move attention shapes to fused_utils · 306de683
  Casper Hansen authored Oct 06, 2023
  
  306de683
- Create ALiBi module · 428504e4
  Casper Hansen authored Oct 06, 2023
  
  428504e4
- Update comment · 7a3d06d6
  Casper Hansen authored Oct 06, 2023
  
  7a3d06d6
- Use new WindowedCache · e46703d8
  Casper Hansen authored Oct 06, 2023
  
  e46703d8
- Create cache for fused modules · 66b2e233
  Casper Hansen authored Oct 06, 2023
  
  66b2e233
05 Oct, 2023 1 commit
- Remove past_key_value (save 2GB VRAM) · 40e6952a
  Casper Hansen authored Oct 05, 2023
  
  40e6952a
02 Oct, 2023 3 commits
- Mistral fused modules · d5bb4ec8
  Casper Hansen authored Oct 02, 2023
  
  d5bb4ec8
- Fix Falcon n_kv_heads parameter · 80eea43e
  Casper Hansen authored Oct 02, 2023
  
  80eea43e
- Fix unexpected keyword · d016c513
  Casper Hansen authored Oct 02, 2023
  
  d016c513
27 Sep, 2023 3 commits
- Fix apply_rotary_emb() to have both tensors on the same device · 841a2313
  s4rduk4r authored Sep 27, 2023
  
  841a2313
- Clean code · cba9a28c
  Casper Hansen authored Sep 27, 2023
  
  cba9a28c
- Remove unused imports · 707315a6
  Casper Hansen authored Sep 27, 2023
  
  707315a6