ayo/ipex-llm
1
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 1
1638 commits 1 branch 0 tags 37 MiB
Commit graph

213 commits

Author SHA1 Message Date
Jiao Wang
aefa5a5bfe Qwen kv cache (#9079) 
* qwen and aquila

* update

* update

* style
 2023-10-05 11:59:17 -07:00
Jiao Wang
d5ca1f32b6 Aquila KV cache optimization (#9080) 
* update

* update

* style
 2023-10-05 11:10:57 -07:00
Yang Wang
88565c76f6 add export merged model example (#9018) 
* add export merged model example

* add sources

* add script

* fix style
 2023-10-04 21:18:52 -07:00
Yang Wang
0cd8f1c79c Use ipex fused rms norm for llama (#9081) 
* also apply rmsnorm

* fix cpu
 2023-10-04 21:04:55 -07:00
Cengguang Zhang
fb883100e7 LLM: support chatglm-18b convert attention forward in benchmark scripts. (#9072) 
* add chatglm-18b convert.

* fix if statement.

* fix
 2023-09-28 14:04:52 +08:00
Yishuo Wang
6de2189e90 [LLM] fix chatglm main choice (#9073) 2023-09-28 11:23:37 +08:00
Cengguang Zhang
b4a1266ef0 [WIP] LLM: add kv cache support for internlm. (#9036) 
* LLM: add kv cache support for internlm

* add internlm apply_rotary_pos_emb

* fix.

* fix style.
 2023-09-25 14:16:59 +08:00
Ruonan Wang
975da86e00 LLM: fix gptneox kv cache (#9044) 2023-09-25 13:03:57 +08:00
Jiao Wang
028a6d9383 MPT model optimize for long sequence (#9020) 
* mpt_long_seq

* update

* update

* update

* style

* style2

* update
 2023-09-21 21:27:23 -07:00
Ruonan Wang
b943d73844 LLM: refactor kv cache (#9030) 
* refactor utils

* meet code review; update all models

* small fix
 2023-09-21 21:28:03 +08:00
Cengguang Zhang
868511cf02 LLM: fix kv cache issue of bloom and falcon. (#9029) 2023-09-21 18:12:20 +08:00
Ruonan Wang
bf51ec40b2 LLM: Fix empty cache (#9024) 
* fix

* fix

* update example
 2023-09-21 17:16:07 +08:00
Yina Chen
714884414e fix error (#9025) 2023-09-21 16:42:11 +08:00
SONG Ge
fa47967583 [LLM] Optimize kv_cache for gptj model family (#9010) 
* optimize gptj model family attention

* add license and comment for dolly-model

* remove xpu mentioned

* remove useless info

* code sytle

* style fix

* code style in gptj fix

* remove gptj arch

* move apply_rotary_pos_emb into utils

* kv_seq_length update

* use hidden_states instead of query layer to reach batch size
 2023-09-21 10:42:08 +08:00
Cengguang Zhang
b3cad7de57 LLM: add bloom kv cache support (#9012) 
* LLM: add bloom kv cache support

* fix style.
 2023-09-20 21:10:53 +08:00
Kai Huang
156af15d1e Add NF3 (#9008) 
* add nf3

* grammar
 2023-09-20 20:03:07 +08:00
Kai Huang
6981745fe4 Optimize kv_cache for gpt-neox model family (#9015) 
* override gptneox

* style

* move to utils

* revert
 2023-09-20 19:59:19 +08:00
Cengguang Zhang
735a17f7b4 LLM: add kv cache to falcon family. (#8995) 
* add kv cache to falcon family.

* fix: import error.

* refactor

* update comments.

* add two version falcon attention forward.

* fix

* fix.

* fix.

* fix.

* fix style.

* fix style.
 2023-09-20 15:36:30 +08:00
Ruonan Wang
94a7f8917b LLM: fix optimized kv cache for baichuan-13b (#9009) 
* fix baichuan 13b

* fix style

* fix

* fix style
 2023-09-20 15:30:14 +08:00
Yang Wang
c88f6ec457 Experiment XPU QLora Finetuning (#8937) 
* Support xpu finetuning

* support xpu finetuning

* fix style

* fix style

* fix style

* refine example

* add readme

* refine readme

* refine api

* fix fp16

* fix example

* refactor

* fix style

* fix compute type

* add qlora

* refine training args

* fix example

* fix style

* fast path forinference

* address comments

* refine readme

* revert lint
 2023-09-19 10:15:44 -07:00
Ruonan Wang
004c45c2be LLM: Support optimized kv_cache for baichuan family (#8997) 
* add initial support for baichuan attantion

* support baichuan1

* update based on comment

* update based on comment

* support baichuan2

* update link, change how to jusge baichuan2

* fix style

* add model parameter for pob emb

* update based on comment
 2023-09-19 15:38:54 +08:00
Zhao Changmin
2a05581da7 LLM: Apply low_cpu_mem_usage algorithm on optimize_model API (#8987) 
* low_cpu_mem_usage
 2023-09-18 21:41:42 +08:00
Zhao Changmin
16b9412e80 tie_word_embeddings (#8977) 
tie_word_embeddings
 2023-09-15 10:17:09 +08:00
Yishuo Wang
bcf456070c fix bloom-176b int overflow (#8973) 2023-09-14 14:37:57 +08:00
Ruonan Wang
dd57623650 LLM: reduce GPU memory for optimize_model=True (#8965) 
* reduce gpu memory for llama & chatglm

* change to device type
 2023-09-13 17:27:09 +08:00
SONG Ge
7132ef6081 [LLM Doc] Add optimize_model doc in transformers api (#8957) 
* add optimize in from_pretrained

* add api doc for load_low_bit

* update api docs following comments

* update api docs

* update

* reord comments
 2023-09-13 10:42:33 +08:00
Zhao Changmin
c32c260ce2 LLM: Add save/load API in optimize_model to support general pytorch model (#8956) 
* support hf format SL
 2023-09-13 10:22:00 +08:00
Guancheng Fu
0bf5857908 [LLM] Integrate FastChat as a serving framework for BigDL-LLM (#8821) 
* Finish changing

* format

* add licence

* Add licence

* fix

* fix

* Add xpu support for fschat

* Fix patch

* Also install webui dependencies

* change setup.py dependency installs

* fiox

* format

* final test
 2023-09-13 09:28:05 +08:00
Zhao Changmin
dcaa4dc130 LLM: Support GQA on llama kvcache (#8938) 
* support GQA
 2023-09-12 12:18:40 +08:00
Yang Wang
16761c58be Make llama attention stateless (#8928) 
* Make llama attention stateless

* fix style

* fix chatglm

* fix chatglm xpu
 2023-09-11 18:21:50 -07:00
Zhao Changmin
e62eda74b8 refine (#8912) 
Co-authored-by: leonardozcm <leonardozcm@gmail.com>
 2023-09-11 16:40:33 +08:00
Yina Chen
df165ad165 init (#8933) 2023-09-11 14:30:55 +08:00
Ruonan Wang
b3f5dd5b5d LLM: update q8 convert xpu&cpu (#8930) 2023-09-08 16:01:17 +08:00
Yina Chen
33d75adadf [LLM]Support q5_0 on arc (#8926) 
* support q5_0

* delete

* fix style
 2023-09-08 15:52:36 +08:00
Yang Wang
ee98cdd85c Support latest transformer version (#8923) 
* Support latest transformer version

* fix style
 2023-09-07 19:01:32 -07:00
Yang Wang
25428b22b4 Fix chatglm2 attention and kv cache (#8924) 
* fix chatglm2 attention

* fix bf16 bug

* make model stateless

* add utils

* cleanup

* fix style
 2023-09-07 18:54:29 -07:00
Yina Chen
b209b8f7b6 [LLM] Fix arc qtype != q4_0 generate issue (#8920) 
* Fix arc precision!=q4_0 generate issue

* meet comments
 2023-09-07 08:56:36 -07:00
Yang Wang
c34400e6b0 Use new layout for xpu qlinear (#8896) 
* use new layout for xpu qlinear

* fix style
 2023-09-06 21:55:33 -07:00
Zhao Changmin
8bc1d8a17c LLM: Fix discards in optimize_model with non-hf models and add openai whisper example (#8877) 
* openai-whisper
 2023-09-07 10:35:59 +08:00
SONG Ge
7a71ced78f [LLM Docs] Remain API Docs Issues Solution (#8780) 
* langchain readthedocs update

* solve langchain.llms.transformersllm issues

* langchain.embeddings.transformersembeddings/transfortmersllms issues

* update docs for get_num_tokens

* add low_bit api doc

* add optimizer model api doc

* update rst index

* fix coomments style

* update docs following the comments

* update api doc
 2023-09-06 16:29:34 +08:00
Kai Huang
4a9ff050a1 Add qlora nf4 (#8782) 
* add nf4

* dequant nf4

* style
 2023-09-06 09:39:22 +08:00
Zhao Changmin
95271f10e0 LLM: Rename low bit layer (#8875) 
* rename lowbit

---------

Co-authored-by: leonardozcm <leonardozcm@gmail.com>
 2023-09-05 13:21:12 +08:00
Yang Wang
242c9d6036 Fix chatglm2 multi-turn streamchat (#8867) 2023-08-31 22:13:49 -07:00
xingyuan li
de6c6bb17f [LLM] Downgrade amx build gcc version and remove avx flag display (#8856) 
* downgrade to gcc 11
* remove avx display
 2023-08-31 14:08:13 +09:00
Yang Wang
3b4f4e1c3d Fix llama attention optimization for XPU (#8855) 
* Fix llama attention optimization fo XPU

* fix chatglm2

* fix typo
 2023-08-30 21:30:49 -07:00
Shengsheng Huang
7b566bf686 [LLM] add new API for optimize any pytorch models (#8827) 
* add new API for optimize any pytorch models

* change test util name

* revise API and update UT

* fix python style

* update ut config, change default value

* change defaults, disable ut transcribe
 2023-08-30 19:41:53 +08:00
Xin Qiu
8eca982301 windows add env (#8852) 2023-08-30 15:54:52 +08:00
Zhao Changmin
731916c639 LLM: Enable attempting loading method automatically (#8841) 
* enable auto load method

* warning error

* logger info

---------

Co-authored-by: leonardozcm <leonardozcm@gmail.com>
 2023-08-30 15:41:55 +08:00
Yishuo Wang
bba73ec9d2 [LLM] change chatglm native int4 checkpoint name (#8851) 2023-08-30 15:05:19 +08:00
Yina Chen
55e705a84c [LLM] Support the rest of AutoXXX classes in Transformers API (#8815) 
* add transformers auto models

* fix
 2023-08-30 11:16:14 +08:00
Yishuo Wang
7429ea0606 [LLM] support transformer int4 + amx int4 (#8838) 2023-08-29 17:27:18 +08:00
Zhao Changmin
bb31d4fe80 LLM: Implement hf low_cpu_mem_usage with 1xbinary file peak memory on transformer int4 (#8731) 
* 1x peak memory
 2023-08-29 09:33:17 +08:00
SONG Ge
d2926c7672 [LLM] Unify Langchain Native and Transformers LLM API (#8752) 
* deprecate BigDLNativeTransformers and add specific LMEmbedding method

* deprecate and add LM methods for langchain llms

* add native params to native langchain

* new imple for embedding

* move ut from bigdlnative to casual llm

* rename embeddings api and examples update align with usage updating

* docqa example hot-fix

* add more api docs

* add langchain ut for starcoder

* support model_kwargs for transformer methods when calling causalLM and add ut

* ut fix for transformers embedding

* update for langchain causal supporting transformers

* remove model_family in readme doc

* add model_families params to support more models

* update api docs and remove chatglm embeddings for now

* remove chatglm embeddings in examples

* new refactor for ut to add bloom and transformers llama ut

* disable llama transformers embedding ut
 2023-08-25 11:14:21 +08:00
Yang Wang
bf3591e2ff Optimize chatglm2 for bf16 (#8725) 
* make chatglm works with bf16

* fix style

* support chatglm v1

* fix style

* fix style

* add chatglm2 file
 2023-08-24 10:04:25 -07:00
Yishuo Wang
611c1fb628 [LLM] change default n_threads of native int4 langchain API (#8779) 2023-08-21 13:30:12 +08:00
Yishuo Wang
3d1f2b44f8 LLM: change default n_threads of native int4 models (#8776) 2023-08-18 15:46:19 +08:00
Yishuo Wang
2ba2133613 fix starcoder chinese output (#8773) 2023-08-18 13:37:02 +08:00
binbin Deng
548f7a6cf7 LLM: update convert of llama family to support llama2-70B (#8747) 2023-08-18 09:30:35 +08:00
Yina Chen
4afea496ab support q8_0 (#8765) 2023-08-17 15:06:36 +08:00
Ruonan Wang
e9aa2bd890 LLM: reduce GPU 1st token latency and update example (#8763) 
* reduce 1st token latency

* update example

* fix

* fix style

* update readme of gpu benchmark
 2023-08-16 18:01:23 +08:00
SONG Ge
f4164e4492 [BigDL LLM] Update readme for unifying transformers API (#8737) 
* update readme doc

* fix readthedocs error

* update comment

* update exception error info

* invalidInputError instead

* fix readme typo error and remove import error

* fix more typo
 2023-08-16 14:22:32 +08:00
Yishuo Wang
77844125f2 [LLM] Support chatglm cache (#8745) 2023-08-14 15:10:46 +08:00
SONG Ge
aceea4dc29 [LLM] Unify Transformers and Native API (#8713) 
* re-open pr to run on latest runner

* re-add examples and ut

* rename ut and move deprecate to warning instead of raising an error info

* ut fix
 2023-08-11 19:45:47 +08:00
Yishuo Wang
f91035c298 [LLM] fix chatglm native int4 emoji output (#8739) 2023-08-11 15:38:41 +08:00
binbin Deng
77efcf7b1d LLM: fix ChatGLM2 native int4 stream output (#8733) 2023-08-11 14:51:50 +08:00
Ruonan Wang
ca3e59a1dc LLM: support stop for starcoder native int4 stream (#8734) 2023-08-11 14:51:30 +08:00
Yishuo Wang
3d5a7484a2 [LLM] fix bloom and starcoder memory release (#8728) 2023-08-11 11:18:19 +08:00
Ruonan Wang
1a7b698a83 [LLM] support ipex arc int4 & add basic llama2 example (#8700) 
* first support of xpu

* make it works on gpu

update setup

update

add GPU llama2 examples

add use_optimize flag to disbale optimize for gpu

fix style

update gpu exmaple readme

fix

* update example, and update env

* fix setup to add cpp files

* replace jit with aot to avoid data leak

* rename to bigdl-core-xe

* update installation in example readme
 2023-08-09 22:20:32 +08:00
Kai Huang
1b65288bdb Add api doc for LLM (#8605) 
* api doc initial

* update desc
 2023-08-08 18:17:16 +08:00
binbin Deng
ea5d7aff5b LLM: add chatglm native int4 transformers API (#8695) 2023-08-07 17:52:47 +08:00
Yishuo Wang
ef08250c21 [LLM] chatglm pybinding support (#8672) 2023-08-04 14:27:29 +08:00
Yang Wang
b6468bac43 optimize chatglm2 long sequence (#8662) 
* add chatglm2

* optimize a little

* optimize chatglm long sequence

* fix style

* address comments and fix style

* fix bug
 2023-08-03 17:56:24 -07:00
Yang Wang
3407f87075 Fix llama kv cache bug (#8674) 2023-08-03 17:54:55 -07:00
binbin Deng
a15a2516e6 add (#8659) 2023-08-03 10:12:10 +08:00
Yina Chen
119bf6d710 [LLM] Support linux cpp dynamic load .so (#8655) 
* support linux cpp dynamic load .so

* update cli
 2023-08-02 20:15:45 +08:00
Zhao Changmin
ca998cc6f2 LLM: Mute shape mismatch output (#8601) 
* LLM: Mute shape mismatch output
 2023-08-02 16:46:22 +08:00
Zhao Changmin
04c713ef06 LLM: Disable transformer api pretraining_tp (#8645) 
* disable pretraining_tp
 2023-08-02 11:26:01 +08:00
Yang Wang
cbeae97a26 Optimize Llama Attention to to reduce KV cache memory copy (#8580) 
* Optimize llama attention to reduce KV cache memory copy

* fix bug

* fix style

* remove git

* fix style

* fix style

* fix style

* fix tests

* move llama attention to another file

* revert

* fix style

* remove jit

* fix
 2023-08-01 16:37:58 -07:00
xingyuan li
cdfbe652ca [LLM] Add chatglm support for llm-cli (#8641) 
* add chatglm build
* add llm-cli support
* update git
* install cmake
* add ut for chatglm
* add files to setup
* fix bug cause permission error when sf lack file
 2023-08-01 14:30:17 +09:00
Zhao Changmin
3e10260c6d LLM: llm-convert support chatglm family (#8643) 
* convert chatglm
 2023-08-01 11:16:18 +08:00
Yina Chen
a607972c0b [LLM]LLM windows load -api.dll (#8631) 
* temp

* update

* revert setup.py
 2023-07-31 13:47:20 +08:00
xingyuan li
3361b66449 [LLM] Revert llm-cli to disable selecting executables on Windows (#8630) 
* revert vnni file select
* revert setup.py
* add model-api.dll
 2023-07-31 11:15:44 +09:00
binbin Deng
fb32fefcbe LLM: support tensor input of native int4 generate (#8620) 2023-07-27 17:59:49 +08:00
Zhao Changmin
5b484ab48d LLM: Support load_low_bit loading models in shards format (#8612) 
* shards_model

---------

Co-authored-by: leonardozcm <leonaordo1997zcm@gmail.com>
 2023-07-26 13:30:01 +08:00
Zhao Changmin
af201052db avoid malloc all missing keys in fp32 (#8600) 2023-07-25 09:48:51 +08:00
Yuwen Hu
ba42a6da63 [LLM] Set torch_dtype default value to 'auto' for transformers low bit from_pretrained API 2023-07-21 17:55:00 +08:00
Yang Wang
feb3af0567 Optimize transformer int4 memory footprint (#8579) 2023-07-20 20:22:13 -07:00
Yang Wang
57e880f63a [LLM] use pytorch linear for large input matrix (#8492) 
* use pytorch linear for large input matrix

* only works on server

* fix style

* optimize memory

* first check server

* revert

* address comments

* fix style
 2023-07-20 09:54:25 -07:00
Zhao Changmin
e680af45ea LLM: Optimize Langchain Pipeline (#8561) 
* LLM: Optimize Langchain Pipeline

* load in low bit
 2023-07-19 17:43:13 +08:00
Zhao Changmin
49d636e295 [LLM] whisper model transformer int4 verification and example (#8511) 
* LLM: transformer api support

* va

* example

* revert

* pep8

* pep8
 2023-07-19 08:33:20 +08:00
Yina Chen
9a7bc17ca1 [LLM] llm supports vnni link on windows (#8543) 
* support win vnni link

* fix style

* fix style

* use isa_checker

* fix

* typo

* fix

* update
 2023-07-18 16:43:45 +08:00
Yina Chen
4582b6939d [LLM]llm gptneox chat (#8527) 
* linux

* support win

* merge upstream & support vnni lib in chat
 2023-07-18 11:17:17 +08:00
Xin Qiu
fccae91461 Add load_low_bit save_load_bit to AutoModelForCausalLM (#8531) 
* transformers save_low_bit load_low_bit

* update example and add readme

* update

* update

* update

* add ut

* update
 2023-07-17 15:29:55 +08:00
xingyuan li
e57db777e0 [LLM] Setup.py & llm-cli update for windows vnni binary files (#8537) 
* update setup.py
* update llm-cli
 2023-07-17 12:28:38 +09:00
Yishuo Wang
6320bf201e LLM: fix memory access violation (#8519) 2023-07-13 17:08:08 +08:00
Xin Qiu
90e3d86bce rename low bit type name (#8512) 
* change qx_0 to sym_intx

* update

* fix typo

* update

* fix type

* fix style

* add python doc

* meet code review

* fix style
 2023-07-13 15:53:31 +08:00
Zhao Changmin
ba0da17b40 LLM: Support AutoModelForSeq2SeqLM transformer API (#8449) 
* LLM: support AutoModelForSeq2SeqLM transformer API
 2023-07-13 13:33:51 +08:00
Yishuo Wang
86b5938075 LLM: fix llm pybinding (#8509) 2023-07-13 10:27:08 +08:00
Zhao Changmin
23f6a4c21f LLM: Optimize transformer int4 loading (#8499) 
* LLM: Optimize transformer int4 loading
 2023-07-12 15:25:42 +08:00
Yishuo Wang
dd3f953288 Support vnni check (#8497) 2023-07-12 10:11:15 +08:00