Commits · 130d5fd8c7902a5a2e54feb25249168151e3bc2f · norm / vllm

04 May, 2023 1 commit
- Fix a bug in attention kernel (#68) · 130d5fd8
  Woosuk Kwon authored May 04, 2023
  
  130d5fd8
03 May, 2023 3 commits
- Support bfloat16 data type (#54) · e070829a
  Woosuk Kwon authored May 03, 2023
  
  e070829a
- Refactor attention kernels (#53) · 436e523b
  Woosuk Kwon authored May 03, 2023
  
  436e523b
- New weight loader without np copy (#52) · 27f1410d
  Zhuohan Li authored May 03, 2023
  
  27f1410d
30 Apr, 2023 1 commit
- Add an option to launch cacheflow without ray (#51) · 4858f3bb
  Zhuohan Li authored Apr 30, 2023
  
  4858f3bb
28 Apr, 2023 1 commit
- Add support for GPT-NeoX (Pythia) (#50) · a96d63c2
  Woosuk Kwon authored Apr 28, 2023
  
  a96d63c2
17 Apr, 2023 1 commit
- Change plotting script · aa50b17c
  Woosuk Kwon authored Apr 17, 2023
  
  aa50b17c
15 Apr, 2023 1 commit
- Support various block sizes & Change default block size to 16 (#38) · 0f4b3219
  Woosuk Kwon authored Apr 15, 2023
  
  0f4b3219
12 Apr, 2023 1 commit
- Collect system stats in scheduler & Add scripts for experiments (#30) · 84eee24e
  Woosuk Kwon authored Apr 12, 2023
  
  84eee24e
11 Apr, 2023 1 commit
- Memcpy kernel for flash attention (#29) · e3cec88a
  Siyuan (Ryans) Zhuang authored Apr 10, 2023
```
* optimize

* add benchmark

* add assert

* add test
```
  e3cec88a
10 Apr, 2023 1 commit
- Support block size 32 (#35) · b9926f7f
  Woosuk Kwon authored Apr 09, 2023
  
  b9926f7f
09 Apr, 2023 1 commit
- Add an option to use dummy model weights (#33) · ee88a7e5
  Woosuk Kwon authored Apr 08, 2023
  
  ee88a7e5
08 Apr, 2023 2 commits
- Add query stride to multi_query_cached_kv_attention & Add kernel benchmark script (#27) · c267b1a0
  Woosuk Kwon authored Apr 08, 2023
```
* Add query stride to multi_query_cached_kv_attention

* Add kernel benchmark script
```
  c267b1a0
- Implement block copy kernel to optimize beam search (#32) · 0f40557a
  Woosuk Kwon authored Apr 07, 2023
  
  0f40557a
06 Apr, 2023 1 commit
- Fix potential bugs in FastAPI frontend and add comments (#28) · a490aafa
  Zhuohan Li authored Apr 06, 2023
  
  a490aafa
05 Apr, 2023 2 commits
- Add CUDA graph-based all reduce launcher (#26) · 12659a0b
  Woosuk Kwon authored Apr 05, 2023
  
  12659a0b
- Basic attention kernel that supports cached KV + (multi-)prompts (#24) · 21b3671b
  Siyuan (Ryans) Zhuang authored Apr 04, 2023
  
  21b3671b
02 Apr, 2023 3 commits
- Optimize data movement (#20) · 897cb2ae
  Woosuk Kwon authored Apr 02, 2023
  
  897cb2ae
- Merge QKV into one linear layer (#15) · 1f01a18d
  Zhuohan Li authored Apr 02, 2023
  
  1f01a18d
- Add ninja to dependency (#21) · 2c5cd0de
  Woosuk Kwon authored Apr 01, 2023
  
  2c5cd0de
01 Apr, 2023 1 commit
- Use FP32 for log probabilities (#19) · a90c97d7
  Woosuk Kwon authored Mar 31, 2023
  
  a90c97d7
31 Mar, 2023 3 commits
- Modify README to include info on loading LLaMA (#18) · e3f00d19
  Zhuohan Li authored Apr 01, 2023
  
  e3f00d19
- Add custom kernel for RMS normalization (#16) · 09e92454
  Woosuk Kwon authored Mar 31, 2023
  
  09e92454
- Optimize tensor parallel execution speed (#17) · c45f3c3a
  Zhuohan Li authored Apr 01, 2023
  
  c45f3c3a
30 Mar, 2023 4 commits
- Implement preemption via recomputation & Refactor scheduling logic (#12) · 7a7929ab
  Woosuk Kwon authored Mar 30, 2023
  
  7a7929ab
- Implement custom kernel for LLaMA rotary embedding (#14) · 88c0268a
  Woosuk Kwon authored Mar 30, 2023
  
  88c0268a
- Implement LLaMA (#9) · 80a2f812
  Woosuk Kwon authored Mar 29, 2023
```
Co-authored-by: Zhuohan Li <zhuohan123@gmail.com>
```
  80a2f812
- Refactor the test code for attention kernels (#13) · a1b3de86
  Woosuk Kwon authored Mar 29, 2023
  
  a1b3de86
29 Mar, 2023 2 commits
- Add cache watermark to avoid frequent cache eviction (#11) · 64e0e383
  Woosuk Kwon authored Mar 29, 2023
  
  64e0e383
- FastAPI-based working frontend (#10) · 721fa3df
  Zhuohan Li authored Mar 29, 2023
  
  721fa3df
26 Mar, 2023 1 commit
- Minor · d359cda5
  Woosuk Kwon authored Mar 26, 2023
  
  d359cda5
21 Mar, 2023 1 commit
- Support tensor parallel (#2) · 2f49f155
  Zhuohan Li authored Mar 22, 2023
  
  2f49f155
13 Mar, 2023 1 commit
- Add miscellaneous updates (#8) · cfae35b8
  Woosuk Kwon authored Mar 13, 2023
  
  cfae35b8
12 Mar, 2023 1 commit
- Add memory analyzer & utomatically configure KV cache size (#6) · e9d3f2ff
  Woosuk Kwon authored Mar 11, 2023
  
  e9d3f2ff
10 Mar, 2023 1 commit
- Support beam search & parallel generation (#7) · 1a7eb7da
  Woosuk Kwon authored Mar 10, 2023
  
  1a7eb7da
06 Mar, 2023 1 commit
- Fix a bug in 1D input shape (#5) · 04e5acc0
  Woosuk Kwon authored Mar 06, 2023
  
  04e5acc0
02 Mar, 2023 1 commit
- Use FlashAttention for `multi_query_kv_attention` (#4) · 3e9f991d
  Woosuk Kwon authored Mar 01, 2023
  
  3e9f991d
01 Mar, 2023 1 commit
- Implement `single_query_cached_kv_attention` kernel (#3) · 0deacbce
  Woosuk Kwon authored Mar 01, 2023
  
  0deacbce
25 Feb, 2023 1 commit
- Fix a bug in tying OPT embeddings (#1) · cbf8779a
  Woosuk Kwon authored Feb 24, 2023
  
  cbf8779a
24 Feb, 2023 1 commit
- Add README · c84c708a
  Woosuk Kwon authored Feb 24, 2023
  
  c84c708a