Huang, Xinshengzi
a2be3d7501
add comment of compress kv in attention forward
2024-08-22 15:11:55 +08:00
Huang, Xinshengzi
ce7de77085
add comment of change in model forward
2024-08-22 14:29:27 +08:00
Huang, Xinshengzi
42398a0045
add comment
2024-08-22 13:17:13 +08:00
Huang, Xinshengzi
48a827aa07
fix typos
2024-08-22 11:35:47 +08:00
Huang, Xinshengzi
8a5df93de2
fix typos
2024-08-22 11:33:07 +08:00
Huang, Xinshengzi
01ed397e7a
fix typos
2024-08-22 11:31:25 +08:00
Huang, Xinshengzi
c6ed1c412d
fix typos
2024-08-22 11:26:49 +08:00
Huang, Xinshengzi
2a0aa9271b
fix typos
2024-08-22 11:23:22 +08:00
Huang, Xinshengzi
4adadddbbc
fix typos
2024-08-22 11:12:23 +08:00
Huang, Xinshengzi
6a5ca17afc
fix typoes
2024-08-22 11:09:58 +08:00
binbin Deng
72a7bf624b
Support qwen2-1.5b with fused decoderlayer optimization on NPU ( #11888 )
2024-08-22 11:09:12 +08:00
Huang, Xinshengzi
6bb9035788
fix typos
2024-08-22 11:08:48 +08:00
Huang, Xinshengzi
86248b0505
add compress_kv for baichuan2
2024-08-22 10:59:08 +08:00
Yina Chen
cc27321441
support chatglm4 in lookup ( #11855 )
2024-08-21 15:53:17 +08:00
Yina Chen
0236de3ac2
set IPEX_LLM_LAST_LM_HEAD=1 as default ( #11885 )
2024-08-21 15:06:12 +08:00
Yang Wang
209d42ab79
Refactor npu mp to make it easier to integrate new models ( #11873 )
...
* Refactor npu mp to make it easier to integrate new models
* fix style
* move layer functions to base
2024-08-20 20:58:47 -07:00
Guancheng Fu
537c0d2767
fix vllm qwen2 models ( #11879 )
2024-08-21 11:05:24 +08:00
Yishuo Wang
bd1e490d62
fix phi3 ( #11878 )
2024-08-21 10:31:41 +08:00
Yang Wang
bdaeee1d63
Fix run_decoders bug ( #11871 )
2024-08-20 12:04:59 -07:00
Yina Chen
c3c058373f
Update compresskv model forward type logic ( #11868 )
...
* update
* fix
2024-08-20 18:11:37 +08:00
Yishuo Wang
d4ee0a89f3
optimize phi3 memory usage ( #11867 )
2024-08-20 17:32:51 +08:00
Yishuo Wang
2946420e14
add minicpmv 2.6 load_low_bit workaround ( #11856 )
2024-08-20 11:16:02 +08:00
Yang Wang
99b05ba1dc
separate prefill into a process ( #11787 )
...
* seperate prefill into a process
* using model.share_memory()
* might work
* worked
* use long prompt
* refactor
* cleanup
* fix bug
* clean up
* changable inter and intra process stages
* refactor
* add max output len
* fix npu_model changes that may cause generate down
* fix npu_model generate import error
* fix generare forward error
---------
Co-authored-by: sgwhat <ge.song@intel.com>
2024-08-19 17:53:36 +08:00
Yishuo Wang
9490781aec
optimize phi3 memory usage again ( #11848 )
2024-08-19 17:26:59 +08:00
Yina Chen
3cd4e87168
Support compress KV with quantize KV ( #11812 )
...
* update llama
* support llama 4.41
* fix style
* support minicpm
* support qwen2
* support minicpm & update
* support chatglm4
* support chatglm
* remove print
* add DynamicCompressFp8Cache & support qwen
* support llama
* support minicpm phi3
* update chatglm2/4
* small fix & support qwen 4.42
* remove print
2024-08-19 15:32:32 +08:00
Zhao Changmin
6841a9ac8f
fix load low bit com dtype ( #11832 )
2024-08-19 13:43:19 +08:00
Yuwen Hu
96796f95cb
Update all-in-one benchmark prompts for continuation task & lookup update for minicpmv ( #11827 )
...
* Update all-in-one benchmark prompts for continuation task
* Small fix
* Add pure-text benchmark support for minicpm-v-2_6
* Support lookahead for model.llm generate of minicpmv
* Add prompt reference
* Small update
* Small fix
2024-08-16 17:16:35 +08:00
Yishuo Wang
e966e85df8
force lm_head optimization in any model if set environment variable ( #11830 )
2024-08-16 16:48:45 +08:00
Yishuo Wang
17a0beb21f
optimize qwen2-audio again ( #11825 )
2024-08-16 11:11:35 +08:00
Yuwen Hu
9e9086cc2a
Update IPEX_LLM_PERFORMANCE_MODE ( #11823 )
2024-08-16 09:48:36 +08:00
Wang, Jian4
5a80fd2633
Fix lightweight-serving no streaming resp on mtl ( #11822 )
2024-08-16 09:43:03 +08:00
Guancheng Fu
e70ae0638e
Fix vLLM not convert issues ( #11817 )
...
* Fix not convert issues
* refine
2024-08-15 19:04:05 +08:00
Yishuo Wang
750d4ad5dc
fix minicpm-v-2 fp16 ( #11819 )
2024-08-15 18:34:40 +08:00
Yishuo Wang
828ab16537
fix phi3 and minicpmv cpu ( #11818 )
2024-08-15 17:43:29 +08:00
Yishuo Wang
4e178f0c5d
rewrite minicpmv optimization ( #11816 )
2024-08-15 17:27:12 +08:00
Yishuo Wang
07b7f13982
support and optimize qwen2-audio ( #11809 )
2024-08-15 14:59:04 +08:00
Yishuo Wang
9a93808fc5
fix and optimize minicpm v 2 ( #11799 )
2024-08-14 17:27:23 +08:00
Yishuo Wang
3d6cfa291d
optimize minicpm v 2.5 ( #11793 )
2024-08-14 16:07:24 +08:00
Yuwen Hu
356281cb80
Further all-in-one benchmark update continuation task ( #11784 )
...
* Further update prompt for continuation task, and disable lookup candidate update strategy on MTL
* style fix
2024-08-14 14:39:34 +08:00
Ruonan Wang
43cca3be27
fix gemma2 runtime error caused by sliding window ( #11788 )
...
* fix runtime error
* revert workflow
2024-08-14 10:43:33 +08:00
Yang Wang
51bcac1229
follow up on experimental support of fused decoder layer for llama2 ( #11785 )
...
* clean up and support transpose value cache
* refine
* fix style
* fix style
2024-08-13 18:53:55 -07:00
Yishuo Wang
cb79dcda93
refactor llama convert to fix minicpm-v 2.5 optimization ( #11783 )
2024-08-14 09:29:57 +08:00
Yina Chen
7cd6ec9723
MiniCPM-V support compresskv ( #11779 )
...
* fix check error
* fix other models
* remove print
2024-08-13 19:03:40 +08:00
Qiyuan Gong
3998de14f0
Fix mistral forward_qkv in q4_0 ( #11781 )
...
* Fix mistral forward_qkv without self.rotary_emb.base in q4_0.
* Replace apply_rotary_pos_emb_no_cache_xpu with rotary_half_inplaced.
* Revert https://github.com/intel-analytics/ipex-llm/pull/11765
2024-08-13 16:48:19 +08:00
Heyang Sun
70c828b87c
deepspeed zero3 QLoRA finetuning ( #11625 )
...
* deepspeed zero3 QLoRA finetuning
* Update convert.py
* Update low_bit_linear.py
* Update utils.py
* Update qlora_finetune_llama2_13b_arch_2_card.sh
* Update low_bit_linear.py
* Update alpaca_qlora_finetuning.py
* Update low_bit_linear.py
* Update utils.py
* Update convert.py
* Update alpaca_qlora_finetuning.py
* Update alpaca_qlora_finetuning.py
* Update low_bit_linear.py
* Update deepspeed_zero3.json
* Update qlora_finetune_llama2_13b_arch_2_card.sh
* Update low_bit_linear.py
* Update low_bit_linear.py
* Update utils.py
* fix style
* fix style
* Update alpaca_qlora_finetuning.py
* Update qlora_finetune_llama2_13b_arch_2_card.sh
* Update convert.py
* Update low_bit_linear.py
* Update model.py
* Update alpaca_qlora_finetuning.py
* Update low_bit_linear.py
* Update low_bit_linear.py
* Update low_bit_linear.py
2024-08-13 16:15:29 +08:00
Yishuo Wang
a184b120c9
fix minicpm-v 2.5 ( #11780 )
2024-08-13 16:14:00 +08:00
Qiyuan Gong
a88c132e54
Reduce Mistral softmax memory only in low memory mode ( #11775 )
...
* Reduce Mistral softmax memory only in low memory mode
2024-08-13 14:50:54 +08:00
Yishuo Wang
aa861df066
use new fp32 softmax kernel ( #11776 )
2024-08-13 14:48:11 +08:00
binbin Deng
23d3acdc77
Add experimental support of fused decoder layer for llama2 ( #11768 )
2024-08-13 14:41:36 +08:00
Yishuo Wang
a1eb793f70
optimize minicpm v 2_6 firs token perf ( #11770 )
2024-08-13 09:51:18 +08:00
Yina Chen
841dbcdf3a
Fix compresskv with lookahead issue ( #11767 )
...
* fix compresskv + lookahead attn_mask qwen2
* support llama chatglm
* support mistral & chatglm
* address comments
* revert run.py
2024-08-12 18:53:55 +08:00
Xu, Shuo
1b05caba2b
Set mistral fuse rope to false except fp6 & fp16 ( #11765 )
...
* set mistral fuse rope to false except fp6 & fp16
* lint
* lint
---------
Co-authored-by: ATMxsp01 <shou.xu@intel.com>
2024-08-12 17:25:07 +08:00
Ruonan Wang
8db34057b4
optimize lookahead init time ( #11769 )
2024-08-12 17:19:12 +08:00
Yishuo Wang
57d177738d
optimize minicpm-v-2_6 repetition penalty ( #11763 )
2024-08-12 14:10:10 +08:00
Wang, Jian4
245dba0abc
Fix lightweight-serving codegeex error ( #11759 )
2024-08-12 10:35:37 +08:00
Ruonan Wang
66fe2ee464
initial support of IPEX_LLM_PERFORMANCE_MODE ( #11754 )
...
* add perf mode
* update
* fix style
2024-08-09 19:04:09 +08:00
Yina Chen
4b9c57cc60
Support compress kv with lookahead ( #11752 )
...
* support compress kv with lookahead
* enough kv miss param
2024-08-09 17:39:57 +08:00
Yishuo Wang
93455aac09
fix minicpm V 2.6 repeat output ( #11753 )
2024-08-09 17:39:24 +08:00
Ruonan Wang
7e917d6cfb
fix gptq of llama ( #11749 )
...
* fix gptq of llama
* small fix
2024-08-09 16:39:25 +08:00
Yina Chen
dd46c141bd
Phi3 support compresskv ( #11733 )
...
* phi3 support compresskv
* fix phi3 mtl error
* fix conflict with quant kv
* fix abnormal on mtl
* fix style
* use slide windows size to compress kv
* support sliding window
* fix style
* fix style
* temp: partial support quant kv
* support quant kv with compress kv, todo: model check
* temp
* fix style
* fix style
* remove prepare
* address comment
* default -> 1.8k
2024-08-09 15:43:43 +08:00
Qiyuan Gong
d8808cc2e3
Mistral apply_rotary_pos_emb_no_cache_xpu use rope_theta from config ( #11747 )
...
mistral-7B-instruct-v0.2 and mistral-7B-instruct-v0.1 use different rope_theta (0.2 is 1e, 0.1 is 1e5). Pass self.config.rope_theta to apply_rotary_pos_emb_no_cache_xpu to avoid output difference.
2024-08-09 10:35:51 +08:00
Xiangyu Tian
044e486480
Fix vLLM CPU /chat endpoint ( #11748 )
2024-08-09 10:33:52 +08:00
Yishuo Wang
54cc9353db
support and optimize minicpm-v-2_6 ( #11738 )
2024-08-07 18:21:16 +08:00
Yina Chen
e956e71fc1
fix conflict with quant kv ( #11737 )
2024-08-07 18:10:30 +08:00
Ruonan Wang
00a5574c8a
Use merge_qkv to replace fused_qkv for llama2 ( #11727 )
...
* update 4.38
* support new versions
* update
* fix style
* fix style
* update rope
* temp test sdpa
* fix style
* fix cpu ut
2024-08-07 18:04:01 +08:00
Yina Chen
d2abc9711b
Fix MTL 4k input qwen2 compresskv error ( #11734 )
...
* fix
* fix style
2024-08-07 16:21:57 +08:00
Yina Chen
a71ae7c22b
Support minicpm compresskv & modify default compresskv config & default enable compresskv on mtl 2.5k~4.5k ( #11726 )
...
* support minicpm & modify default & default enable on mtl 2.5k~4.5k
* fix style
2024-08-07 11:35:39 +08:00
Yishuo Wang
c093f7d980
fix phi3 ( #11729 )
2024-08-07 09:39:46 +08:00
Zijie Li
e7f7141781
Add benchmark util for transformers 4.42 ( #11725 )
...
* add new benchmark_util.py
Add new benchmark_util.py for transformers>=4.43.1. The old one renamed to benchmark_util_prev.py.
* Small fix to import code
* Update __init__.py
* fix file names
* Update lint-python
Update lint-python to exclude benchmark_util_4_29.py
benchmark_util_4_43.py
* Update benchmark_util_4_43.py
* add benchmark_util for transformers 4.42
2024-08-07 08:48:07 +08:00
Yishuo Wang
929675aa6b
support latest phi3 ( #11721 )
2024-08-06 15:52:55 +08:00
Yishuo Wang
bbdff6edeb
optimize internvl2 4b performance ( #11720 )
2024-08-06 14:25:08 +08:00
Yishuo Wang
f44b732aa8
support internvl2-4b ( #11718 )
2024-08-06 13:36:32 +08:00
Zijie Li
8fb36b9f4a
add new benchmark_util.py ( #11713 )
...
* add new benchmark_util.py
2024-08-05 16:18:48 +08:00
Wang, Jian4
493cbd9a36
Support lightweight-serving with internlm-xcomposer2-vl-7b multimodal input ( #11703 )
...
* init image_list
* enable internlm-xcomposer2 image input
* update style
* add readme
* update model
* update readme
2024-08-05 09:36:04 +08:00
Ruonan Wang
aa98ef96fe
change mixed_precision to q6_k ( #11706 )
2024-08-02 15:55:16 +08:00
Xiangyu Tian
1baa3efe0e
Optimizations for Pipeline Parallel Serving ( #11702 )
...
Optimizations for Pipeline Parallel Serving
2024-08-02 12:06:59 +08:00
Yina Chen
8d1e0bd2f4
add sdp causal support in llama ( #11705 )
2024-08-02 10:27:40 +08:00
Ruonan Wang
736a7ef72e
add sdp_causal for mistral 4.36 ( #11686 )
...
* add sdp_causal for mistral
* fix
* update
2024-08-01 18:57:31 +08:00
Yina Chen
45c730ff39
Chatglm support compresskv ( #11690 )
...
* chatglm4 support compresskv
* fix
* fix style
* support chatglm2
* fix quantkv conflict
* fix style
2024-08-01 18:20:20 +08:00
Guancheng Fu
afeca38a47
Fix import vllm condition ( #11682 )
2024-07-31 13:50:01 +08:00
Ruonan Wang
54bf3a23a6
add fallback for unsupported k-quants ( #11691 )
...
* add fallback
* fix style
* fix
2024-07-31 11:39:58 +08:00
Wang, Jian4
b119825152
Remove tgi parameter validation ( #11688 )
...
* remove validation
* add min warm up
* remove no need source
2024-07-30 16:37:44 +08:00
Yina Chen
670ad887fc
Qwen support compress kv ( #11680 )
...
* Qwen support compress kv
* fix style
* fix
2024-07-30 11:16:42 +08:00
hxsz1997
9b36877897
disable default quantize_kv of GQA on MTL ( #11679 )
...
* disable default quantizekv of gqa in mtl
* fix stype
* fix stype
* fix stype
* fix stype
* fix stype
* fix stype
2024-07-30 09:38:46 +08:00
Yishuo Wang
c02003925b
add mlp for gemma2 ( #11678 )
2024-07-29 16:10:23 +08:00
Yishuo Wang
6f999e6e90
add sdp for gemma2 ( #11677 )
2024-07-29 15:15:47 +08:00
Ruonan Wang
c11d5301d7
add sdp fp8 for llama ( #11671 )
...
* add sdp fp8 for llama
* fix style
* refactor
2024-07-29 13:46:22 +08:00
Yishuo Wang
7f88ce23cd
add more gemma2 optimization ( #11673 )
2024-07-29 11:13:00 +08:00
Yishuo Wang
3e8819734b
add basic gemma2 optimization ( #11672 )
2024-07-29 10:46:51 +08:00
Heyang Sun
ba01b85c13
empty cache only for 1st token but rest token to speed up ( #11665 )
2024-07-26 16:46:21 +08:00
Yina Chen
fc7f8feb83
Support compress kv ( #11642 )
...
* mistral snapkv
* update
* mtl update
* update
* update
* update
* add comments
* style fix
* fix style
* support llama
* llama use compress kv
* support mistral 4.40
* fix style
* support diff transformers versions
* move snapkv util to kv
* fix style
* meet comments & small fix
* revert all in one
* fix indent
---------
Co-authored-by: leonardozcm <leonardo1997zcm@gmail.com>
2024-07-26 16:02:00 +08:00
Yishuo Wang
6bcdc6cc8f
fix qwen2 cpu ( #11663 )
2024-07-26 13:41:51 +08:00
Wang, Jian4
23681fbf5c
Support codegeex4-9b for lightweight-serving ( #11648 )
...
* add options, support prompt and not return end_token
* enable openai parameter
* set do_sample None and update style
2024-07-26 09:41:03 +08:00
Guancheng Fu
a4d30a8211
Change logic for detecting if vllm is available ( #11657 )
...
* fix
* fix
2024-07-25 15:24:19 +08:00
Xiangyu Tian
4499d25c26
LLM: Fix ParallelLMHead convert in vLLM cpu ( #11654 )
2024-07-25 13:07:19 +08:00
binbin Deng
777e61d8c8
Fix qwen2 & int4 on NPU ( #11646 )
2024-07-24 13:14:39 +08:00
Yishuo Wang
1b3b46e54d
fix chatglm new model ( #11639 )
2024-07-23 13:44:56 +08:00
Xiangyu Tian
060792a648
LLM: Refine Pipeline Parallel FastAPI ( #11587 )
...
Refine Pipeline Parallel FastAPI
2024-07-22 15:52:05 +08:00
Wang, Jian4
1eed0635f2
Add lightweight serving and support tgi parameter ( #11600 )
...
* init tgi request
* update openai api
* update for pp
* update and add readme
* add to docker
* add start bash
* update
* update
* update
2024-07-19 13:15:56 +08:00
Xiangyu Tian
d27a8cd08c
Fix Pipeline Parallel dtype ( #11623 )
2024-07-19 13:07:40 +08:00
Yishuo Wang
d020ad6397
add save_low_bit support for DiskEmbedding ( #11621 )
2024-07-19 10:34:53 +08:00
Guoqiong Song
380717f50d
fix gemma for 4.41 ( #11531 )
...
* fix gemma for 4.41
2024-07-18 15:02:50 -07:00
Guoqiong Song
5a6211fd56
fix minicpm for transformers>=4.39 ( #11533 )
...
* fix minicpm for transformers>=4.39
2024-07-18 15:01:57 -07:00
Yishuo Wang
0209427cf4
Add disk_embedding parameter to support put Embedding layer on CPU ( #11617 )
2024-07-18 17:06:06 +08:00
Xiangyu Tian
4594a3dd6c
LLM: Fix DummyLayer.weight device in Pipeline Parallel ( #11612 )
2024-07-18 13:39:34 +08:00
Yishuo Wang
f4077fa905
fix llama3-8b npu long input stuck ( #11613 )
2024-07-18 11:08:17 +08:00
Zhao Changmin
e5c0058c0e
fix baichuan ( #11606 )
2024-07-18 09:43:36 +08:00
Guoqiong Song
d64711900a
Fix cohere model on transformers>=4.41 ( #11575 )
...
* fix cohere model for 4-41
2024-07-17 17:18:59 -07:00
Wang, Jian4
9c15abf825
Refactor fastapi-serving and add one card serving( #11581 )
...
* init fastapi-serving one card
* mv api code to source
* update worker
* update for style-check
* add worker
* update bash
* update
* update worker name and add readme
* rename update
* rename to fastapi
2024-07-17 11:12:43 +08:00
Yishuo Wang
5837bc0014
fix chatglm3 npu output ( #11590 )
2024-07-16 18:16:30 +08:00
Guancheng Fu
06930ab258
Enable ipex-llm optimization for lm head ( #11589 )
...
* basic
* Modify convert.py
* fix
2024-07-16 16:48:44 +08:00
Yina Chen
99c22745b2
fix qwen 14b fp6 abnormal output ( #11583 )
2024-07-16 10:59:00 +08:00
Yishuo Wang
c279849d27
add disk embedding api ( #11585 )
2024-07-16 10:43:39 +08:00
Xiangyu Tian
79c742dfd5
LLM: Add XPU Memory Optimizations for Pipeline Parallel ( #11567 )
...
Add XPU Memory Optimizations for Pipeline Parallel
2024-07-16 09:44:50 +08:00
Yishuo Wang
019da6c0ab
use mlp silu_mul fusion in qwen2 to optimize memory usage ( #11574 )
2024-07-13 16:32:54 +08:00
Yishuo Wang
a945500a98
fix internlm xcomposser stream chat ( #11564 )
2024-07-11 18:21:17 +08:00
Zhao Changmin
b9c66994a5
add npu sdp ( #11562 )
2024-07-11 16:57:35 +08:00
binbin Deng
2b8ad8731e
Support pipeline parallel for glm-4v ( #11545 )
2024-07-11 16:06:06 +08:00
Xiangyu Tian
7f5111a998
LLM: Refine start script for Pipeline Parallel Serving ( #11557 )
...
Refine start script and readme for Pipeline Parallel Serving
2024-07-11 15:45:27 +08:00
Zhao Changmin
105e124752
optimize phi3-v encoder npu performance and add multimodal example ( #11553 )
...
* phi3-v
* readme
2024-07-11 13:59:14 +08:00
Cengguang Zhang
70ab1a6f1a
LLM: unify memory optimization env variables. ( #11549 )
...
* LLM: unify memory optimization env variables.
* fix comments.
2024-07-11 11:01:28 +08:00
Yishuo Wang
994e49a510
optimize internlm xcomposser performance again ( #11551 )
2024-07-10 17:08:56 +08:00
Yishuo Wang
82f9514303
optimize internlm xcomposer2 performance ( #11550 )
2024-07-10 15:57:04 +08:00
Zhao Changmin
3c16c9f725
Optimize baichuan on NPU ( #11548 )
...
* baichuan_npu
2024-07-10 13:18:48 +08:00
Yishuo Wang
7dc6756d86
add disk embedding ( #11543 )
2024-07-09 17:38:40 +08:00
Yishuo Wang
99b2802d3b
optimize qewn2 memory ( #11535 )
2024-07-09 17:14:01 +08:00
Yishuo Wang
2929eb262e
support npu glm4 ( #11539 )
2024-07-09 15:46:49 +08:00
Xiangyu Tian
a1cede926d
Fix update_kv_cache in Pipeline-Parallel-Serving for glm4-9b model ( #11537 )
2024-07-09 14:08:04 +08:00
Yishuo Wang
c26651f91f
add mistral npu support ( #11523 )
2024-07-08 13:17:15 +08:00
binbin Deng
252426793b
Fix setting of use_quantize_kv_cache on different GPU in pipeline parallel ( #11516 )
2024-07-08 09:27:01 +08:00
Yishuo Wang
7cb09a8eac
optimize qwen2 memory usage again ( #11520 )
2024-07-05 17:32:34 +08:00
Zhao Changmin
f7e957aaf9
Clean npu dtype branch ( #11515 )
...
* clean branch
* create_npu_kernels
2024-07-05 15:45:26 +08:00
Yishuo Wang
14ce058004
add chatglm3 npu support ( #11518 )
2024-07-05 15:31:27 +08:00
Xin Qiu
a31f2cbe13
update minicpm.py ( #11517 )
...
* update minicpm
* meet code review
2024-07-05 15:25:44 +08:00
Zhao Changmin
24de13fc45
Optimize stablelm on NPU ( #11512 )
...
* stablelm_optimize
2024-07-05 14:21:57 +08:00
Xiangyu Tian
7d8bc83415
LLM: Partial Prefilling for Pipeline Parallel Serving ( #11457 )
...
LLM: Partial Prefilling for Pipeline Parallel Serving
2024-07-05 13:10:35 +08:00
binbin Deng
60de428b37
Support pipeline parallel for qwen-vl ( #11503 )
2024-07-04 18:03:57 +08:00
Zhao Changmin
57b8adb189
[WIP] Support npu load_low_bit method ( #11502 )
...
* npu_load_low_bit
2024-07-04 17:15:34 +08:00
Yishuo Wang
1a8bab172e
add minicpm 1B/2B npu support ( #11507 )
2024-07-04 16:31:04 +08:00
Yishuo Wang
bb0a84044b
add qwen2 npu support ( #11504 )
2024-07-04 11:01:25 +08:00
Xin Qiu
f84ca99b9f
optimize gemma2 rmsnorm ( #11500 )
2024-07-03 15:21:03 +08:00
binbin Deng
9274282ef7
Support pipeline parallel for glm-4-9b-chat ( #11463 )
2024-07-03 14:25:28 +08:00
Yishuo Wang
d97c2664ce
use new fuse rope in stablelm family ( #11497 )
2024-07-03 11:08:26 +08:00
Zhao Changmin
6a0134a9b2
support q4_0_rtn ( #11477 )
...
* q4_0_rtn
2024-07-02 16:57:02 +08:00
Yishuo Wang
5e967205ac
remove the code converts input to fp16 before calling batch forward kernel ( #11489 )
2024-07-02 16:23:53 +08:00
Yishuo Wang
ec3a912ab6
optimize npu llama long context performance ( #11478 )
2024-07-01 16:49:23 +08:00
binbin Deng
48ad482d3d
Fix import error caused by pydantic on cpu ( #11474 )
2024-07-01 15:49:49 +08:00
Yishuo Wang
39bcb33a67
add sdp support for stablelm 3b ( #11473 )
2024-07-01 14:56:15 +08:00
Zhao Changmin
cf8eb7b128
Init NPU quantize method and support q8_0_rtn ( #11452 )
...
* q8_0_rtn
* fix float point
2024-07-01 13:45:07 +08:00
Yishuo Wang
319a3b36b2
fix npu llama2 ( #11471 )
2024-07-01 10:14:11 +08:00
SONG Ge
a414e3ff8a
add pipeline parallel support with load_low_bit ( #11414 )
2024-06-28 10:17:56 +08:00
binbin Deng
987017ef47
Update pipeline parallel serving for more model support ( #11428 )
2024-06-27 18:21:01 +08:00
Yishuo Wang
029ff15d28
optimize npu llama2 first token performance ( #11451 )
2024-06-27 17:37:33 +08:00
Qiyuan Gong
4e4ecd5095
Control sys.modules ipex duplicate check with BIGDL_CHECK_DUPLICATE_IMPORT ( #11453 )
...
* Control sys.modules ipex duplicate check with BIGDL_CHECK_DUPLICATE_IMPORT。
2024-06-27 17:21:45 +08:00
Yishuo Wang
c6e5ad668d
fix internlm xcomposser meta-instruction typo ( #11448 )
2024-06-27 15:29:43 +08:00
Yishuo Wang
f89ca23748
optimize npu llama2 perf again ( #11445 )
2024-06-27 15:13:42 +08:00
Yishuo Wang
2a0f8087e3
optimize qwen2 gpu memory usage again ( #11435 )
2024-06-26 16:52:29 +08:00
Shaojun Liu
ab9f7f3ac5
FIX: Qwen1.5-GPTQ-Int4 inference error ( #11432 )
...
* merge_qkv if quant_method is 'gptq'
* fix python style checks
* refactor
* update GPU example
2024-06-26 15:36:22 +08:00
Guancheng Fu
99cd16ef9f
Fix error while using pipeline parallism ( #11434 )
2024-06-26 15:33:47 +08:00
Yishuo Wang
ca0e69c3a7
optimize npu llama perf again ( #11431 )
2024-06-26 10:52:54 +08:00
Yishuo Wang
9f6e5b4fba
optimize llama npu perf ( #11426 )
2024-06-25 17:43:20 +08:00
binbin Deng
e473b8d946
Add more qwen1.5 and qwen2 support for pipeline parallel inference ( #11423 )
2024-06-25 15:49:32 +08:00
binbin Deng
aacc1fd8c0
Fix shape error when run qwen1.5-14b using deepspeed autotp ( #11420 )
2024-06-25 13:48:37 +08:00
Xiangyu Tian
8ddae22cfb
LLM: Refactor Pipeline-Parallel-FastAPI example ( #11319 )
...
Initially Refactor for Pipeline-Parallel-FastAPI example
2024-06-25 13:30:36 +08:00
Xin Qiu
9e4ee61737
rename BIGDL_OPTIMIZE_LM_HEAD to IPEX_LLM_LAST_LM_HEAD and add qwen2 ( #11418 )
2024-06-24 18:42:37 +08:00
Yishuo Wang
abe53eaa4f
optimize qwen1.5/2 memory usage when running long input with fp16 ( #11403 )
2024-06-24 13:43:04 +08:00
Guoqiong Song
7507000ef2
Fix 1383 Llama model on transformers=4.41[WIP] ( #11280 )
2024-06-21 11:24:10 -07:00
SONG Ge
0c67639539
Add more examples for pipeline parallel inference ( #11372 )
...
* add more model exampels for pipelien parallel inference
* add mixtral and vicuna models
* add yi model and past_kv supprot for chatglm family
* add docs
* doc update
* add license
* update
2024-06-21 17:55:16 +08:00
Xiangyu Tian
b30bf7648e
Fix vLLM CPU api_server params ( #11384 )
2024-06-21 13:00:06 +08:00
binbin Deng
4ba82191f2
Support PP inference for chatglm3 ( #11375 )
2024-06-21 09:59:01 +08:00
Yishuo Wang
f0fdfa081b
Optimize qwen 1.5 14B batch performance ( #11370 )
2024-06-20 17:23:39 +08:00
Yishuo Wang
a5e7d93242
Add initial save/load low bit support for NPU(now only fp16 is supported) ( #11359 )
2024-06-20 10:49:39 +08:00
Qiyuan Gong
1eb884a249
IPEX Duplicate importer V2 ( #11310 )
...
* Add gguf support.
* Avoid error when import ipex-llm for multiple times.
* Add check to avoid duplicate replace and revert.
* Add calling from check to avoid raising exceptions in the submodule.
* Add BIGDL_CHECK_DUPLICATE_IMPORT for controlling duplicate checker. Default is true.
2024-06-19 16:29:19 +08:00
Yishuo Wang
ae7b662ed2
add fp16 NPU Linear support and fix intel_npu_acceleration_library version 1.0 support ( #11352 )
2024-06-19 09:14:59 +08:00
Guoqiong Song
c44b1942ed
fix mistral for transformers>=4.39 ( #11191 )
...
* fix mistral for transformers>=4.39
2024-06-18 13:39:35 -07:00
Yishuo Wang
83082e5cc7
add initial support for intel npu acceleration library ( #11347 )
2024-06-18 16:07:16 +08:00
Yina Chen
5dad33e5af
Support fp8_e4m3 scale search ( #11339 )
...
* fp8e4m3 switch off
* fix style
2024-06-18 11:47:43 +08:00
binbin Deng
e50c890e1f
Support finishing PP inference once eos_token_id is found ( #11336 )
2024-06-18 09:55:40 +08:00
SONG Ge
ef4b6519fb
Add phi-3 model support for pipeline parallel inference ( #11334 )
...
* add phi-3 model support
* add phi3 example
2024-06-17 17:44:24 +08:00
Xin Qiu
183e0c6cf5
glm-4v-9b support ( #11327 )
...
* chatglm4v support
* fix style check
* update glm4v
2024-06-17 13:52:37 +08:00
binbin Deng
6ea1e71af0
Update PP inference benchmark script ( #11323 )
2024-06-17 09:59:36 +08:00
SONG Ge
be00380f1a
Fix pipeline parallel inference past_key_value error in Baichuan ( #11318 )
...
* fix past_key_value error
* add baichuan2 example
* fix style
* update doc
* add script link in doc
* fix import error
* update
2024-06-17 09:29:32 +08:00
Yina Chen
0af0102e61
Add quantization scale search switch ( #11326 )
...
* add scale_search switch
* remove llama3 instruct
* remove print
2024-06-14 18:46:52 +08:00
Ruonan Wang
8a3247ac71
support batch forward for q4_k, q6_k ( #11325 )
2024-06-14 18:25:50 +08:00
Yishuo Wang
e8dd8e97ef
fix chatglm lookahead on ARC ( #11320 )
2024-06-14 16:26:11 +08:00
Yishuo Wang
91965b5d05
add glm_sdpa back to fix chatglm-6b ( #11313 )
2024-06-14 10:31:43 +08:00
Yishuo Wang
7f65836cb9
fix chatglm2/3-32k/128k fp16 ( #11311 )
2024-06-14 09:58:07 +08:00
Xin Qiu
1b0c4c8cb8
use new rotary two in chatglm4 ( #11312 )
...
* use new rotary two in chatglm4
* rempve
2024-06-13 19:02:18 +08:00
Xin Qiu
f1410d6823
refactor chatglm4 ( #11301 )
...
* glm4
* remove useless code
* stype
* add rope_ratio
* update
* fix fp16
* fix style
2024-06-13 18:06:04 +08:00
Yishuo Wang
5e25766855
fix and optimize chatglm2-32k and chatglm3-128k ( #11306 )
2024-06-13 17:37:58 +08:00
binbin Deng
60cb1dac7c
Support PP for qwen1.5 ( #11300 )
2024-06-13 17:35:24 +08:00
Yishuo Wang
a24666b8f3
fix chatglm3-6b-32k ( #11303 )
2024-06-13 16:01:34 +08:00
Yishuo Wang
01fe0fc1a2
refactor chatglm2/3 ( #11290 )
2024-06-13 12:22:58 +08:00
Guancheng Fu
57a023aadc
Fix vllm tp ( #11297 )
2024-06-13 10:47:48 +08:00
binbin Deng
220151e2a1
Refactor pipeline parallel multi-stage implementation ( #11286 )
2024-06-13 10:00:23 +08:00
Ruonan Wang
14b1e6b699
Fix gguf_q4k ( #11293 )
...
* udpate embedding parameter
* update benchmark
2024-06-12 20:43:08 +08:00
Yuwen Hu
8edcdeb0e7
Fix bug that torch.ops.torch_ipex.matmul_bias_out cannot work on Linux MTL for short input ( #11292 )
2024-06-12 19:12:57 +08:00
Xin Qiu
592f7aa61e
Refine glm1-4 sdp ( #11276 )
...
* chatglm
* update
* update
* change chatglm
* update sdpa
* update
* fix style
* fix
* fix glm
* update glm2-32k
* update glm2-32k
* fix cpu
* update
* change lower_bound
2024-06-12 17:11:56 +08:00
Yuwen Hu
cffb932f05
Expose timeout for streamer for fastchat worker ( #11288 )
...
* Expose timeout for stremer for fastchat worker
* Change to read from env variables
2024-06-12 17:02:40 +08:00
Qiyuan Gong
0d9cc9c106
Remove duplicate check for ipex ( #11281 )
...
* Replacing builtin.import is causing lots of unpredicted problems. Remove this function.
2024-06-12 13:52:02 +08:00
