468d761b32
[Misc] Reduce supported Punica dtypes ( #4304 )
2024-04-23 18:54:33 -07:00
91f50a6fe2
[Core][Distributed] use cpu/gloo to initialize pynccl ( #4248 )
2024-04-23 18:32:19 -07:00
1e8f4252aa
[Bugfix][Frontend] Raise exception when file-like chat template fails to be opened ( #4292 )
2024-04-23 18:19:03 +00:00
2b7949c1c2
AQLM CUDA support ( #3287 )
...
Co-authored-by: mgoin <michael@neuralmagic.com >
2024-04-23 13:59:33 -04:00
62b8aebc6f
[Speculative decoding 7/9] Speculative decoding end-to-end correctness tests. ( #3951 )
2024-04-23 08:02:36 +00:00
050f285ff6
[Core] Scheduling optimization 2 ( #4280 )
2024-04-23 08:02:11 +00:00
ad8d696a99
[Core] Scheduler perf fix ( #4270 )
2024-04-22 21:11:06 +00:00
a37d815b83
Make initialization of tokenizer and detokenizer optional ( #3748 )
...
Co-authored-by: Yun Ding <yunding@nvidia.com >
Co-authored-by: Roger Wang <ywang@roblox.com >
2024-04-21 22:06:46 +00:00
91528575ec
[Frontend] multiple sampling params support ( #3570 )
2024-04-20 00:11:57 -07:00
a22cdea371
[Kernel][FP8] Initial support with dynamic per-tensor scaling ( #4118 )
...
Provide an initial support to FP8 computation. This PR is inspired by HuggingFace TGI: huggingface/text-generation-inference#1726
This feature can be enabled with --quantization fp8 or -q fp8 when launching an engine.
Algorithm:
We still load a model checkpoint in FP16/BF16. After the weights are loaded, Fp8LinearMethod calculates the per-tensor scaling factor of weights and quantizes the weights accordingly. The scaling factor will then be stored for future use. Meanwhile, the per-tensor scaling factor for activations is calculated in every forward pass.
Initial Results:
Currently tested Mistral-7B on 1xH100. With prompt length ~5 and decoding length 128:
BF16: 1.47s
FP8: 1.66s
I'll try to use larger models and try to find more performance bottleneck. Meanwhile, you're welcome to try this code.
2024-04-20 04:28:57 +00:00
138485a82d
[Bugfix] Add fix for JSON whitespace ( #4189 )
...
Co-authored-by: Ubuntu <ubuntu@ip-172-31-13-147.ec2.internal >
2024-04-19 20:49:22 -07:00
d17c8477f1
[Bugfix] Fix LoRA loading check ( #4138 )
...
Co-authored-by: simon-mo <simon.mo@hey.com >
2024-04-19 00:59:54 -07:00
8a7a3e4436
[Core] add an option to log every function call to for debugging hang/crash in distributed inference ( #4079 )
...
Co-authored-by: Simon Mo <simon.mo@hey.com >
2024-04-18 16:15:12 -07:00
e1bb2fd52d
[Bugfix] Support logprobs when using guided_json and other constrained decoding fields ( #4149 )
2024-04-18 21:12:55 +00:00
e8cc7967ff
[Bugfix][Kernel] allow non-power-of-two head sizes in prefix prefill ( #4128 )
2024-04-18 00:51:28 -07:00
53b018edcb
[Bugfix] Get available quantization methods from quantization registry ( #4098 )
2024-04-18 00:21:55 -07:00
6dc1fc9cfe
[Core] nccl integrity check and test ( #4155 )
...
[Core] Add integrity check during initialization; add test for it (#4155 )
2024-04-17 22:28:52 -07:00
a53222544c
[Kernel] Add punica dimension for Swallow-MS-7B LoRA ( #4134 )
2024-04-17 10:02:45 -07:00
8438e0569e
[Core] RayWorkerVllm --> WorkerWrapper to reduce duplication ( #4024 )
...
[Core] replace narrow-usage RayWorkerVllm to general WorkerWrapper to reduce code duplication (#4024 )
2024-04-17 08:34:33 +00:00
e95cd87959
[Speculative decoding 6/9] Integrate speculative decoding with LLMEngine ( #3894 )
2024-04-16 13:09:21 -07:00
69e1d2fb69
[Core] Refactor model loading code ( #4097 )
2024-04-16 11:34:39 -07:00
05434764cd
LM Format Enforcer Guided Decoding Support ( #3868 )
...
Co-authored-by: Simon Mo <simon.mo@hey.com >
2024-04-16 05:54:57 +00:00
4e7ee664e2
[Core] Fix engine-use-ray broken ( #4105 )
2024-04-16 05:24:53 +00:00
711a000255
[Frontend] [Core] feat: Add model loading using tensorizer ( #3476 )
2024-04-13 17:13:01 -07:00
989ae2538d
[Kernel] Add punica dimension for Baichuan-13B ( #4053 )
2024-04-13 07:55:05 -07:00
36729bac13
[Test] Test multiple attn backend for chunked prefill. ( #4023 )
2024-04-12 09:56:57 -07:00
1096717ae9
[Core] Support LoRA on quantized models ( #4012 )
2024-04-11 21:02:44 -07:00
e46a60aa4c
[BugFix] Fix handling of stop strings and stop token ids ( #3672 )
2024-04-11 15:34:12 -07:00
1e96c3341a
Add extra punica sizes to support bigger vocabs ( #4015 )
2024-04-11 22:18:57 +00:00
95e7d4a97c
Fix echo/logprob OpenAI completion bug ( #3441 )
...
Co-authored-by: Dylan Hawk <dylanwawk@gmail.com >
2024-04-11 22:15:50 +00:00
a10d3056da
[Core] Set linear_weights directly on the layer ( #3977 )
2024-04-11 16:35:51 -04:00
e9da5a40c6
[Misc] Add indirection layer for custom ops ( #3913 )
2024-04-10 20:26:07 -07:00
e42df7227d
[Test] Add xformer and flash attn tests ( #3961 )
...
Co-authored-by: Simon Mo <simon.mo@hey.com >
2024-04-11 03:09:50 +00:00
67b4221a61
[Core][5/N] Fully working chunked prefill e2e ( #3884 )
2024-04-10 17:56:48 -07:00
63e7176f26
[Core][Refactor] move parallel_utils into vllm/distributed ( #3950 )
...
[WIP][Core][Refactor] move vllm/model_executor/parallel_utils into vllm/distributed and vllm/device_communicators (#3950 )
2024-04-10 15:33:30 -07:00
0258b7a94b
[Bugfix] handle prompt_logprobs in _apply_min_tokens_penalty ( #3876 )
...
Signed-off-by: Travis Johnson <tsjohnso@us.ibm.com >
2024-04-10 01:39:56 -07:00
b3104b2a10
[Bugfix] Fix logits processor when prompt_logprobs is not None ( #3899 )
2024-04-10 00:09:36 -07:00
11dd6ebb89
[Misc] Avoid loading incorrect LoRA config ( #3777 )
2024-04-09 19:47:15 -07:00
e7c7067b45
[Misc] [Core] Implement RFC "Augment BaseExecutor interfaces to enable hardware-agnostic speculative decoding" ( #3837 )
2024-04-09 11:44:15 -07:00
95baec828f
[Core] enable out-of-tree model register ( #3871 )
2024-04-06 17:11:41 -07:00
18de883489
[Chunked Prefill][4/n] Chunked prefill scheduler. ( #3853 )
2024-04-05 10:17:58 -07:00
e5043a3e75
[Misc] Add pytest marker to opt-out of global test cleanup ( #3863 )
2024-04-04 21:54:16 -07:00
aabe8f40f2
[Core] [Frontend] Make detokenization optional ( #3749 )
...
Co-authored-by: Nick Hill <nickhill@us.ibm.com >
2024-04-03 21:52:18 -07:00
537ee25f43
[Core] Enable hf_transfer by default if available ( #3817 )
2024-04-04 04:02:43 +00:00
2ff767b513
Enable scaled FP8 (e4m3fn) KV cache on ROCm (AMD GPU) ( #3290 )
...
Co-authored-by: Gregory Shtrasberg <Gregory.Shtrasberg@amd.com >
Co-authored-by: HaiShaw <hixiao@gmail.com >
Co-authored-by: AdrianAbeyta <Adrian.Abeyta@amd.com >
Co-authored-by: Matthew Wong <Matthew.Wong2@amd.com >
Co-authored-by: root <root@gt-pla-u18-08.pla.dcgpu >
Co-authored-by: mawong-amd <156021403+mawong-amd@users.noreply.github.com >
Co-authored-by: ttbachyinsda <ttbachyinsda@outlook.com >
Co-authored-by: guofangze <guofangze@kuaishou.com >
Co-authored-by: Michael Goin <mgoin64@gmail.com >
Co-authored-by: jacobthebanana <50071502+jacobthebanana@users.noreply.github.com >
Co-authored-by: Woosuk Kwon <woosuk.kwon@berkeley.edu >
2024-04-03 14:15:55 -07:00
3dcb3e8b98
[3/N] Refactor scheduler for chunked prefill scheduling ( #3550 )
2024-04-03 14:13:49 -07:00
5757d90e26
[Speculative decoding] Adding configuration object for speculative decoding ( #3706 )
...
Co-authored-by: Lily Liu <lilyliupku@gmail.com >
2024-04-03 00:40:57 +00:00
eb69d68804
[Misc] [CI/Build] Speed up block manager CPU-only unit tests ~10x by opting-out of GPU cleanup ( #3783 )
2024-04-02 00:49:51 +00:00
7d4e1b85e7
[Misc] Add support for new autogptq checkpoint_format ( #3689 )
...
Co-authored-by: Robert Shaw <rshaw@neuralmagic.com >
2024-04-01 19:32:01 -04:00
93deb0b38f
[Speculative decoding 4/9] Lookahead scheduling for speculative decoding ( #3250 )
2024-04-01 22:55:24 +00:00
