Lianmin Zheng
|
dfec7fca06
|
Rename sglang.bench_latency to sglang.bench_one_batch (#2118)
|
2024-11-21 20:07:48 -08:00 |
|
Jake Poznanski
|
8048c28c11
|
Fix #2037 - Context length check does not take into out pad tokens for visual models (#2106)
|
2024-11-21 19:05:41 -08:00 |
|
Byron Hsu
|
30af7dfb34
|
[router] add base_gpu_id server args & merged radix tree python reference (#2115)
|
2024-11-21 17:13:33 -08:00 |
|
James Xu
|
f6f713797b
|
Add support for Qwen2-VL-based embedding models (#2055)
|
2024-11-21 14:24:25 -08:00 |
|
HAI
|
f35cb46cc3
|
ROCm: Fix MoE padding for none FP8 cases (#2111)
|
2024-11-21 12:23:21 -08:00 |
|
Jerry Zhang
|
7f8fcd39cd
|
Turn off autotune for scaled mm for fp8 dynamic quant in torchao (#2116)
|
2024-11-21 12:19:49 -08:00 |
|
Jerry Zhang
|
5c6a41facf
|
Error out when torchao-config option is not recognized (#2107)
|
2024-11-20 17:37:28 -08:00 |
|
Lianmin Zheng
|
722530fa01
|
Enable overlap scheduler by default for the triton attention backend (#2105)
|
2024-11-20 02:58:35 -08:00 |
|
Lianmin Zheng
|
3295cd8af2
|
Allow skipping warmup in bench_offline_throughput.py (#2103)
|
2024-11-20 01:25:21 -08:00 |
|
Ying Sheng
|
5942dfc00a
|
[feat] Add session control (#2073)
|
2024-11-20 00:36:53 -08:00 |
|
Lianmin Zheng
|
7d671e4ad2
|
Enable overlap by default (#2067)
|
2024-11-19 22:07:58 -08:00 |
|
Ke Bao
|
699384cb01
|
Set schedule policy more conservative for DP attention (#2096)
|
2024-11-19 20:57:18 -08:00 |
|
Lianmin Zheng
|
ffd20fcd03
|
Make constrained decoding work for overlap scheduler (#2095)
|
2024-11-19 15:04:43 -08:00 |
|
Yineng Zhang
|
55bd97f3e5
|
minor: add dataset dump and questions shuffle (#2093)
|
2024-11-19 14:07:27 -08:00 |
|
HAI
|
e57c3e12b8
|
Use native fp8 format on MI300X (#2094)
|
2024-11-19 14:06:29 -08:00 |
|
Alexander Waitz
|
929c7621af
|
Fix: incorrect top_logprobs in chat completion (#2088)
|
2024-11-19 12:21:36 +00:00 |
|
Lianmin Zheng
|
b7a065eae3
|
Use cuda event wait and synchronization instead of busy waiting (#2089)
|
2024-11-19 00:21:46 -08:00 |
|
Lianmin Zheng
|
b110453802
|
Simplify logits penalizer (#2086)
|
2024-11-18 17:48:28 -08:00 |
|
Lianmin Zheng
|
3b44bbeecf
|
Allow passing extra request body to bench_offline_throughput.py (#2085)
|
2024-11-18 14:59:15 -08:00 |
|
Lianmin Zheng
|
80e2c4a8de
|
Fix chunked prefill with output logprob (#2083)
|
2024-11-18 13:16:28 -08:00 |
|
Jani Monoses
|
66318ffe96
|
Rename layer_idx to layer_id for consistency (#2078)
|
2024-11-18 13:00:02 -08:00 |
|
Yineng Zhang
|
766192610e
|
feat: update torch 2.5.1 (#2069)
|
2024-11-18 21:29:13 +08:00 |
|
yukavio
|
2a3992b6f1
|
support set role as 'tool' (#2075)
Co-authored-by: kavioyu <kavioyu@tencent.com>
|
2024-11-18 01:06:59 -08:00 |
|
Lianmin Zheng
|
4af3f889fc
|
Simplify flashinfer indices update for prefill (#2074)
Co-authored-by: kavioyu <kavioyu@tencent.com>
Co-authored-by: kavioyu <kavioyu@gmail.com>
|
2024-11-18 00:02:36 -08:00 |
|
Lianmin Zheng
|
df7fe4521a
|
Crash the CI jobs on model import errors (#2072)
|
2024-11-17 22:18:11 -08:00 |
|
Lianmin Zheng
|
116685337e
|
Fix cuda illegal memory access in overlap mode (#2070)
|
2024-11-17 21:29:30 -08:00 |
|
Lianmin Zheng
|
a9e90b4bce
|
[Minor] Fix styles for overlap mode (#2068)
|
2024-11-17 19:49:20 -08:00 |
|
Tanjiro
|
8c280cee55
|
add phi-3 small support (#2062)
Co-authored-by: Tushar Goel <114812108+AI-Tushar@users.noreply.github.com>
|
2024-11-17 18:47:43 -08:00 |
|
DarkSharpness
|
9c745d078e
|
[Performance] Update xgrammar-related constrained decoding (#2056)
|
2024-11-17 16:58:49 -08:00 |
|
Lianmin Zheng
|
ebaa2f3199
|
Rename arguments --disable-nan-detection to --enable-nan-detection (#2066)
|
2024-11-17 16:53:44 -08:00 |
|
Ke Bao
|
62832bb272
|
Support cuda graph for DP attention (#2061)
|
2024-11-17 16:29:20 -08:00 |
|
Lianmin Zheng
|
11f881d173
|
Deprecate --disable-flashinfer and --disable-flashinfer-sampling (#2065)
|
2024-11-17 16:20:58 -08:00 |
|
Lianmin Zheng
|
38625e2139
|
Remove monkey_patch_vllm_dummy_weight_loader (#2064)
|
2024-11-17 15:48:12 -08:00 |
|
Lianmin Zheng
|
c1f401fc58
|
Revert "chore: update torch v2.5.1" (#2063)
|
2024-11-17 15:29:38 -08:00 |
|
Yineng Zhang
|
3b878863f7
|
chore: update torch v2.5.1 (#1849)
|
2024-11-18 00:06:00 +08:00 |
|
Lianmin Zheng
|
f719d9aebc
|
Launch dp ranks in parallel (#2053)
Co-authored-by: Haotian Liu <6631389+haotian-liu@users.noreply.github.com>
|
2024-11-16 17:39:39 -08:00 |
|
Lianmin Zheng
|
edad373135
|
Fix illegal memory access in overlap mode & Use more fused triton kernels for building meta data (#2051)
|
2024-11-16 16:14:23 -08:00 |
|
Ke Bao
|
976bc302e5
|
Support DP MLA (#1970)
|
2024-11-16 09:01:43 +00:00 |
|
Lianmin Zheng
|
2f2e07439c
|
Fix weight update for data parallelism (#2050)
|
2024-11-16 00:30:39 -08:00 |
|
HAI
|
2ffe0a7363
|
Add get_amdgpu_memory_capacity() (#2049)
|
2024-11-15 22:51:48 -08:00 |
|
Ke Wen
|
cf2489762b
|
Add Tensor Parallel to torch_native_llama (#1876)
|
2024-11-15 21:26:00 -08:00 |
|
HAI
|
e5c6715003
|
Fix core (MI300X) with --enable-overlap (#2048)
|
2024-11-15 21:24:42 -08:00 |
|
Lianmin Zheng
|
32c9a7ec11
|
Release v0.3.5.post2 (#2046)
|
2024-11-15 06:54:00 -08:00 |
|
Lianmin Zheng
|
b01df48cf2
|
[Fix] Adjust default chunked prefill size and cuda graph max bs according to GPU memory capacity (#2044)
|
2024-11-15 06:21:57 -08:00 |
|
Lianmin Zheng
|
c29b98e043
|
Fix json benchmark (#2043)
|
2024-11-15 05:33:43 -08:00 |
|
Lianmin Zheng
|
2558d6a675
|
Fix the default arguments of bench_offline_throughput.py & simplify detokenizer manager (#2042)
|
2024-11-15 05:02:44 -08:00 |
|
ws
|
29ebe3dff4
|
fix: align enable_overlap_scheduler naming between code and docs (#2038)
|
2024-11-15 03:39:10 -08:00 |
|
zolinthecow
|
f6dd648620
|
Offline LLM Engine Benchmark Throughput (#1968)
Co-authored-by: ByronHsu <byronhsu1230@gmail.com>
|
2024-11-14 21:59:33 -08:00 |
|
Lianmin Zheng
|
ea53c63bad
|
Expose no_stop_trim and skip_special_tokens in openai api (#2039)
|
2024-11-14 19:09:21 -08:00 |
|
Lianmin Zheng
|
a10d530943
|
Fix outlines version (#2036)
|
2024-11-14 12:52:40 -08:00 |
|