Commit Graph
1439 Commits
Author SHA1 Message Date
Liangsheng Yin bb3a3b6675 Support Faster JSON decoding for llava (#137)
When sending fast-forwarded reqs to model_rpc, re-calculate `pad_input_ids`
2024-02-03 23:32:05 +08:00
Ying Sheng 45d6592d40 Fix no-cache mode (#136) 2024-02-03 04:59:06 -08:00
Ying Sheng e095b16236 Add max_prefill_num_token into server arguments (#133) 2024-02-03 02:35:54 -08:00
Lianmin Zheng c7af9f7393 Fix a bug in llava-hd 2024-01-31 18:52:15 +00:00
Lianmin Zheng ad82bac6f5 Fix model loading & format code (#125) 2024-01-30 23:49:52 -08:00
Cody Yu 71b54eea7d Add cache metrics (#119) 2024-01-30 22:13:14 -08:00
Lianmin Zheng 74b3bfaaf8 format code 2024-01-30 16:36:10 +00:00
Lianmin Zheng 873d0e8537 Ignore detokenization error 2024-01-30 14:52:06 +00:00
Keith Stevens 1d0fbe8e43 [Feature] Adds basic support for image content in OpenAI chat routes (#113) 2024-01-30 06:12:33 -08:00
Lianmin Zheng 97aa9b3284 Improve docs & Add JSON decode example (#121) 2024-01-30 05:45:27 -08:00
Lianmin Zheng 6f560c761b Improve the control of streaming and improve the first token latency in streaming (#117) 2024-01-29 17:05:42 -08:00
Liangsheng Yin 81561f8e2d Flush Cache API (#103) 2024-01-25 21:32:59 -08:00
Cody Yu 3a581e9949 Dynamic model class loading (#101) 2024-01-25 15:29:07 -08:00
parasol-aserandYing Sheng 23950056f0 support speculative execution for openai API (#48)
Co-authored-by: Ying Sheng <sqy1415@gmail.com>
2024-01-25 01:57:06 -08:00
Liangsheng Yin 01ee0fbc05 fast regex decode
Auto-detect constant str path in regex FSM, then extend instead.
2024-01-25 01:16:25 +08:00
Lianmin Zheng c70b3cfa9e Bump the version to v0.1.8 (#93) 2024-01-24 03:33:34 -08:00
Ying Sheng 489796c7ea minor performance fix 2024-01-24 10:45:44 +00:00
Lianmin Zheng fa7a696d04 Fix max_new_tokens for limited memory 2024-01-24 10:44:32 +00:00
Lianmin Zheng bef0b35902 Fix llava & Fix multiprocessing 2024-01-24 10:35:31 +00:00
shiyi.c_98andHaotian Liu c6576e820c Llava-hd Support (#92)
Co-authored-by: Haotian Liu <liuhaotian.cn@gmail.com>
2024-01-24 01:51:21 -08:00
Lianmin Zheng 9a16fea012 Return logprob for choices (#87) 2024-01-23 05:07:30 -08:00
Lianmin Zheng 94e05770db Fix after QWen support (#82) 2024-01-22 21:17:05 -08:00
Arcmoon 63e97e5e4c Suppport qwen model and solve some problems (#75) 2024-01-22 20:14:51 -08:00
Lianmin Zheng a837166e6f Fix select and normalized logprobs (#67) 2024-01-21 01:39:23 -08:00
Lianmin Zheng 11f3cca64f Fix select (#64) 2024-01-20 23:20:35 -08:00
Liangsheng Yin ca13f3b8c5 Disk FSM cache and adjust code. (#63) 2024-01-20 21:26:11 -08:00
Lianmin Zheng f30abd090a Improve error message & Add vicuna template (#57) 2024-01-19 17:03:33 -08:00
Liangsheng Yin 40ab1f0129 Fix the possible bug of decode out of memory (#36) 2024-01-19 11:01:15 -08:00
Cody Yu 23471f9aa3 Support v1/chat/completions (#50) 2024-01-18 23:43:09 -08:00
Cody Yu 61d4c93962 Support stream=True in v1/completions (#49) 2024-01-18 17:00:56 -08:00
Lianmin Zheng 22ec7bc2a1 Expose more arguments to control the scheduling policy (#32) 2024-01-17 18:37:02 -08:00
Lianmin Zheng f9d723816a Teak mem fraction (#20) 2024-01-17 04:43:17 -08:00
Lianmin Zheng bf51ddc6e5 Improve docs & Rename Gemini -> VertexAI (#19) 2024-01-17 02:54:41 -08:00
Lianmin Zheng c4707f1bb5 Improve docs (#17) 2024-01-16 19:53:55 -08:00
Ying ShengandLianmin Zheng ffe4aaee1d Fix for T4 GPUs (#16)
Co-authored-by: Lianmin Zheng <lianminzheng@gmail.com>
2024-01-16 15:49:03 -08:00
Liangsheng Yin 01ca82d765 fix radix cache match (#7) 2024-01-15 09:42:46 -08:00
Lianmin Zheng 4bd8233f2c Fix test cases (#6) 2024-01-15 01:15:53 -08:00
Liangsheng Yin 08ab2a1655 Json Decode && Mutl-Turns (#4) 2024-01-15 00:49:29 -08:00
22085081bb release initial code
Co-authored-by: Ying Sheng <sqy1415@gmail.com>
Co-authored-by: Liangsheng Yin <hnyls2002@gmail.com>
Co-authored-by: Zhiqiang Xie <xiezhq@stanford.edu>
Co-authored-by: parasol-aser <3848358+parasol-aser@users.noreply.github.com>
Co-authored-by: LiviaSun <33578456+ChuyueSun@users.noreply.github.com>
Co-authored-by: Cody Yu <hao.yu.cody@gmail.com>
2024-01-08 04:37:50 +00:00