Wang, Jian4
c8a0462507
Add vllm api_server input output log ( #12962 )
2025-03-12 20:58:04 +08:00
Yishuo Wang
b6f33d5c4d
optimize moonlight again ( #12909 )
2025-03-03 09:21:15 +08:00
Yishuo Wang
39e360fe9d
add grouped topk optimization for moonlight ( #12903 )
2025-02-28 13:25:56 +08:00
Xin Qiu
e946127613
glm 4v 1st sdp for vision ( #12904 )
...
* glm4v 1st sdp
* update glm4v example
* meet code review
* fix style
2025-02-28 13:23:27 +08:00
Yishuo Wang
be1f073866
add fuse moe optimization for moonlight ( #12898 )
2025-02-27 09:15:24 +08:00
Yishuo Wang
5faba06409
simple optimization for moonlight moe decoding forward ( #12891 )
2025-02-25 16:18:27 +08:00
Yishuo Wang
ab3fc66eb7
optimize attention part of moonlight-14B-A3B ( #12886 )
2025-02-25 09:38:13 +08:00
Yishuo Wang
3f6ecce508
support using xgrammar to get json output ( #12870 )
2025-02-24 14:10:58 +08:00
Wang, Jian4
3ea5389a99
Fix vllm api_server v1/models error ( #12867 )
2025-02-21 11:08:29 +08:00
Wang, Jian4
348dc8056d
Fix vllm gptq awq error ( #12863 )
...
* fix gptq awq error
* fix python style
2025-02-20 16:27:23 +08:00
Guancheng Fu
4eed0c7d99
initial implementation for low_bit_loader vLLM ( #12838 )
...
* initial
* add logic for handling tensor parallel models
* fix
* Add some comments
* add doc
* fix done
2025-02-19 19:45:34 +08:00
Xiangyu Tian
b26409d53f
R1 Hybrid: Add Benchmark for DeepSeek R1 transformers example ( #12854 )
...
* init
* fix
* update
* update
* fix
* fix
2025-02-19 18:33:21 +08:00
Yishuo Wang
aee2db30f9
update sdp support ( #12847 )
2025-02-19 12:07:00 +08:00
Xiangyu Tian
93c10be762
LLM: Support hybrid convert for DeepSeek V3/R1 ( #12834 )
...
LLM: Support hybrid convert for DeepSeek V3/R1
2025-02-19 11:31:19 +08:00
Wang, Jian4
e1809a6295
Update multimodal on vllm 0.6.6 ( #12816 )
...
* add glm4v and minicpmv example
* fix
2025-02-19 10:04:42 +08:00
Yishuo Wang
8418450300
optimize minicpm-o's tts part ( #12833 )
2025-02-17 14:53:37 +08:00
Wang, Jian4
1083fe5508
Reenable pp and lightweight-serving serving on 0.6.6 ( #12814 )
...
* reenable pp ang lightweight serving on 066
* update readme
* updat
* update tag
2025-02-13 10:16:00 +08:00
Guancheng Fu
af693425f1
Upgrade to vLLM 0.6.6 ( #12796 )
...
* init
* update engine init
* fix serving load_in_low_bit problem
* temp
* temp
* temp
* temp
* temp
* fix
* fixed
* done
* fix
* fix all arguments
* fix
* fix throughput script
* fix
* fix
* use official ipex-llm
* Fix readme
* fix
---------
Co-authored-by: hzjane <a1015616934@qq.com>
2025-02-12 16:47:51 +08:00
Yishuo Wang
f8ab833f74
support and optimize janus pro ( #12813 )
2025-02-12 15:07:24 +08:00
Yishuo Wang
73cfe293fa
add basic support for Baichuan-M1-14B-Instruct ( #12808 )
2025-02-11 17:27:42 +08:00
Xiangyu Tian
b70ad902b4
Fix ipex-llm CPU linear dtype not match ( #12805 )
2025-02-11 10:34:44 +08:00
Yishuo Wang
e4ceb722b6
fix qwen2 vl ( #12798 )
2025-02-10 13:25:53 +08:00
Ruonan Wang
e90a9ad196
[NPU] Support non-const parameter for decoder layers when keep_ir=True ( #12789 )
...
* support layernorm=False for decoder layers
* renbame to meet review
* fix style
* rename to const_parameter
* fix rebase error
* fix rebase error
2025-02-08 09:58:42 +08:00
binbin Deng
ca1d7b7c2c
[NPU] Support qwen models with cos_sin_input=True ( #12788 )
2025-02-07 16:41:13 +08:00
Yishuo Wang
d0d9c9d636
remove load_in_8bit usage as it is not supported a long time ago ( #12779 )
2025-02-07 11:21:29 +08:00
Yishuo Wang
9697197f3e
fix qlora finetune example ( #12769 )
2025-02-06 11:18:28 +08:00
Ruonan Wang
094a25b740
[NPU] Expose parameter to control blob / IR save logic ( #12767 )
...
* update api
* fix convert.py
* fix style
* remove unnecessary bin file
* fix style
2025-02-06 10:07:45 +08:00
Yishuo Wang
0237ffb302
refactor xpu linear forward ( #12768 )
2025-02-05 17:40:38 +08:00
Yuwen Hu
69f13c78b8
[NPU] Update layernorm node on MTL/ARL ( #12738 )
...
* Update layernorm node on MTL/ARL
* Fix on style
2025-01-23 17:25:19 +08:00
Yuwen Hu
dcca522618
Remove sdpa available patch ( #12734 )
2025-01-22 17:22:28 +08:00
Xiangyu Tian
c9b6c94a59
vLLM: Update vLLM-cpu to v0.6.6-post1 ( #12728 )
...
Update vLLM-cpu to v0.6.6-post1
2025-01-22 15:03:01 +08:00
Yishuo Wang
6789e5d92f
small fix ( #12727 )
2025-01-21 17:27:18 +08:00
Yishuo Wang
085974e307
fix nf4 to cpu ( #12722 )
2025-01-21 09:23:22 +08:00
Yuwen Hu
9aa4be8ced
Update runtime configuration on MTL ( #12720 )
2025-01-20 11:06:37 +08:00
Yishuo Wang
bda87c21eb
add support and optimization for minicpmo audio part ( #12716 )
2025-01-16 16:39:00 +08:00
Zhao Changmin
54d6328b3c
woq int4 fwd ( #12711 )
2025-01-16 15:48:05 +08:00
Yishuo Wang
b62734748f
add support and optimization for minicpmo vision part ( #12713 )
2025-01-16 14:51:00 +08:00
Yuwen Hu
9d65dcd7ef
Fix deepseek coder with linear rope type support on GPU ( #12709 )
...
* Fix deepseek coder with linear rope type
* Style fix
* Move to optimize_pre
* Small fix
* Small fix
* Small fix to not affect other cases
* Style fixes
* Update function name
* Small fix
* Small fix
* Small fix
* Fix for low transformers version first
* Style fix
* Small fix
2025-01-15 21:12:34 +08:00
Cengguang Zhang
9930351112
LLM: add new qtype woq_int4 to support gemm int4 temporary. ( #12706 )
...
This PR add temporary qtype woq_int4 to avoid affecting other qtype and models.
Co-authored-by: leonardozcm <leonardo1997zcm@gmail.com>
2025-01-15 14:41:33 +08:00
Xu, Shuo
350fae285d
Add Qwen2-VL HF GPU example with ModelScope Support ( #12606 )
...
* Add qwen2-vl example
* complete generate.py & readme
* improve lint style
* update 1-6
* update main readme
* Format and other small fixes
---------
Co-authored-by: Yuwen Hu <yuwen.hu@intel.com>
2025-01-13 15:42:04 +08:00
Yuwen Hu
a1da7908b9
Fix name device is not found bug ( #12703 )
2025-01-13 10:11:02 +08:00
Yishuo Wang
db9db51e2c
fix lnl perf ( #12700 )
2025-01-10 18:00:58 +08:00
binbin Deng
da8bcb7db1
[NPU ] fix load logic of glm-edge models ( #12698 )
2025-01-10 16:08:37 +08:00
Yishuo Wang
f8dc408888
fix user issue ( #12692 )
2025-01-10 10:18:47 +08:00
Yishuo Wang
68857494a5
refactor to simplify following upgrade 2 ( #12685 )
2025-01-10 09:29:03 +08:00
Yishuo Wang
7234c9b27b
update quantize kv cache condition ( #12681 )
2025-01-09 15:23:04 +08:00
Yishuo Wang
1ec40cd09e
refactor to simplify following upgrade ( #12680 )
2025-01-09 13:34:30 +08:00
Yishuo Wang
5c24276fc4
fix custom kernel registration ( #12674 )
2025-01-08 17:39:17 +08:00
Yishuo Wang
a22a8c21bb
small fix and remove ununsed code about ipex ( #12671 )
2025-01-08 17:39:04 +08:00
Yishuo Wang
c11f5f0fcd
also convert SdpaAttention in optimize_model ( #12673 )
2025-01-08 16:48:03 +08:00
