LMDeploy Release V0.3.0

Highlight

• Refactor attention and optimize GQA(#1258 #1307 #1116), achieving 22+ and 16+ RPS for internlm2-7b and internlm2-20b, about 1.8x faster than vLLM
• Support new models, including Qwen1.5-MOE(#1372), DBRX(#1367), DeepSeek-VL(#1335)

What's Changed

🚀 Features

• Add tensor core GQA dispatch for [4,5,6,8] by @lzhangzz in #1258
• upgrade turbomind to v2.1 by by @lzhangzz in #1307, #1116
• Support slora to pipeline by @AllentDan in #1286
• Support qwen for pytorch engine by @RunningLeon in #1265
• Support Triton inference server python backend by @ispobock in #1329
• torch engine support dbrx by @grimoire in #1367
• Support qwen2 moe for pytorch engine by @RunningLeon in #1372
• Add deepseek vl by @AllentDan in #1335

💥 Improvements

• rm unused var by @zhyncs in #1256
• Expose cache_block_seq_len to API by @ispobock in #1218
• add chat template for deepseek coder model by @lvhan028 in #1310
• Add more log info for api_server by @AllentDan in #1323
• remove cuda cache after loading vison model by @irexyc in #1325
• Add new chat cli with auto backend feature by @RunningLeon in #1276
• Update rewritings for qwen by @RunningLeon in #1351
• lazy import accelerate.init_empty_weights for vl async engine by @irexyc in #1359
• update lmdeploy pypi packages deps to cuda12 by @irexyc in #1368
• update max_prefill_token_num for low gpu memory by @grimoire in #1373
• Optimize pipeline of pytorch engine by @grimoire in #1328

🐞 Bug fixes

• fix different stop/bad words length in batch by @irexyc in #1246
• Fix performance issue of chatbot by @ispobock in #1295
• add missed argument by @irexyc in #1317
• Fix dlpack memory leak by @ispobock in #1344
• Fix invalid context for Internstudio platform by @lzhangzz in #1354
• fix benchmark generation by @grimoire in #1349
• fix window attention by @grimoire in #1341
• fix batchApplyRepetitionPenalty by @irexyc in #1358
• Fix memory leak of DLManagedTensor by @ispobock in #1361
• fix vlm inference hung with tp by @irexyc in #1336
• [Fix] fix the unit test of model name deduce by @AllentDan in #1382

📚 Documentations

• add citation in readme by @RunningLeon in #1308
• Add slora example for pipeline by @AllentDan in #1343

🌐 Other

• Add restful interface regrssion daily test workflow. by @zhulinJulia24 in #1302
• Add offline mode for testcase workflow by @zhulinJulia24 in #1318
• workflow bugfix and add llava-v1.5-13b testcase by @zhulinJulia24 in #1339
• Add benchmark test workflow by @zhulinJulia24 in #1364
• bump version to v0.3.0 by @lvhan028 in #1387

Full Changelog: v0.2.6...v0.3.0

LMDeploy Release V0.2.6

Highlight

Support vision-languange models (VLM) inference pipeline and serving.
Currently, it supports the following models, Qwen-VL-Chat, LLaVA series v1.5, v1.6 and Yi-VL

• VLM Inference Pipeline

from lmdeploy import pipeline
from lmdeploy.vl import load_image

pipe = pipeline('liuhaotian/llava-v1.6-vicuna-7b')

image = load_image('https://raw.githubusercontent.com/open-mmlab/mmdeploy/main/tests/data/tiger.jpeg')
response = pipe(('describe this image', image))
print(response)

Please refer to the detailed guide from here

• VLM serving by openai compatible server

lmdeploy server api_server liuhaotian/llava-v1.6-vicuna-7b --server-port 8000

• VLM Serving by gradio

lmdeploy serve gradio liuhaotian/llava-v1.6-vicuna-7b --server-port 6006

What's Changed

🚀 Features

• Add inference pipeline for VL models by @irexyc in #1214
• Support serving VLMs by @AllentDan in #1285
• Serve VLM by gradio by @irexyc in #1293
• Add pipeline.chat api for easy use by @irexyc in #1292

💥 Improvements

• Hide qos functions from swagger UI if not applied by @AllentDan in #1238
• Color log formatter by @grimoire in #1247
• optimize filling kv cache kernel in pytorch engine by @grimoire in #1251
• Refactor chat template and support accurate name matching. by @AllentDan in #1216
• Support passing json file to chat template by @AllentDan in #1200
• upgrade peft and check adapters by @grimoire in #1284
• better cache allocation in pytorch engine by @grimoire in #1272
• Fall back to base template if there is no chat_template in tokenizer_config.json by @AllentDan in #1294

🐞 Bug fixes

• lazy load convert_pv jit function by @grimoire in #1253
• [BUG] fix the case when num_used_blocks < 0 by @jjjjohnson in #1277
• Check bf16 model in torch engine by @grimoire in #1270
• fix bf16 check by @grimoire in #1281
• [Fix] fix triton server chatbot init error by @AllentDan in #1278
• Fix concatenate issue in profile serving by @ispobock in #1282
• fix torch tp lora adapter by @grimoire in #1300
• Fix crash when api_server loads a turbomind model by @irexyc in #1304

📚 Documentations

• fix config for readthedocs by @RunningLeon in #1245
• update badges in README by @lvhan028 in #1243
• Update serving guide including api_server and gradio by @lvhan028 in #1248
• rename restful_api.md to api_server.md by @lvhan028 in #1287
• Update readthedocs index by @lvhan028 in #1288

🌐 Other

• Parallelize testcase and refactor test workflow by @zhulinJulia24 in #1254
• Accelerate sample request in benchmark script by @ispobock in #1264
• Update eval ci cfg by @RunningLeon in #1259
• Test case bugfix and add restful interface testcases. by @zhulinJulia24 in #1271
• bump version to v0.2.6 by @lvhan028 in #1299

New Contributors

• @jjjjohnson made their first contribution in #1277

Full Changelog: v0.2.5...v0.2.6

LMDeploy Release V0.2.5

What's Changed

🚀 Features

• Support mistral and sliding window attention by @grimoire in #1075
• torch engine support chatglm3 by @grimoire in #1159
• Support qwen1.5 in pytorch engine by @grimoire in #1160
• Support mixtral for pytorch engine by @RunningLeon in #1133
• Support torch deepseek moe by @grimoire in #1163
• Support gemma model in pytorch engine by @grimoire in #1184
• Auto backend for pipeline and serve when backend is not set to pytorch explicitly by @RunningLeon in #1211

💥 Improvements

• Fix argument error by @ispobock in #1193
• Use LifoQueue for turbomind async_stream_infer by @AllentDan in #1179
• Update interactive output len strategy and response by @AllentDan in #1164
• Support min_new_tokens generation config in pytorch engine by @grimoire in #1096
• Batched sampling by @grimoire in #1197
• refactor the logic of getting model_name by @AllentDan in #1188
• Add parameter max_prefill_token_num by @lvhan028 in #1203
• optmize baichuan in pytorch engine by @grimoire in #1223
• check model required transformers version by @grimoire in #1220
• torch optmize chatglm3 by @grimoire in #1215
• Async torch engine by @grimoire in #1206
• remove unused kernel in pytorch engine by @grimoire in #1237

🐞 Bug fixes

• Fix session length for profile generation by @ispobock in #1181
• fix torch engine infer by @RunningLeon in #1185
• fix module map by @grimoire in #1205
• [Fix] Correct session length warning by @AllentDan in #1207
• Fix all devices occupation when applying tp to torch engine by updating device map by @grimoire in #1172
• Fix falcon chatglm2 template by @grimoire in #1168
• [Fix] Avoid AsyncEngine running the same session id by @AllentDan in #1219
• Fix None session_len by @lvhan028 in #1230
• fix multinomial sampling by @grimoire in #1228
• fix returning logits in prefill phase of pytorch engine by @grimoire in #1209
• optimize pytorch engine inference with falcon model by @grimoire in #1234
• fix bf16 multinomial sampling by @grimoire in #1239
• reduce torchengine prefill mem usage by @grimoire in #1240

📚 Documentations

• auto generate pipeline api for readthedocs by @RunningLeon in #1186
• Added tutorial document for deploying lmdeploy on Jetson series boards. by @BestAnHongjun in #1192
• update doc index by @zhyncs in #1241

🌐 Other

• Add PR test workflow and check-in more testcases by @zhulinJulia24 in #1208
• fix pytest version by @zhulinJulia24 in #1236
• bump version to v0.2.5 by @lvhan028 in #1235

New Contributors

• @ispobock made their first contribution in #1181
• @BestAnHongjun made their first contribution in #1192

Full Changelog: v0.2.4...v0.2.5

LMDeploy Release V0.2.4

What's Changed

💥 Improvements

• use stricter rules to get weight file by @irexyc in #1070
• check pytorch engine environment by @grimoire in #1107
• Update Dockerfile order to launch the http service by docker run directly by @AllentDan in #1162
• Support torch cache_max_entry_count by @grimoire in #1166
• Remove the manual model conversion during benchmark by @lvhan028 in #953
• update llama triton example by @zhyncs in #1153

🐞 Bug fixes

• fix embedding copy size by @irexyc in #1036
• fix pytorch engine with peft==0.8.2 by @grimoire in #1122
• support triton2.2 by @grimoire in #1137
• Add top_k in ChatCompletionRequest by @lvhan028 in #1174
• minor fix benchmark generation guide and script by @lvhan028 in #1175

📚 Documentations

• docs add debug turbomind guide by @zhyncs in #1121

🌐 Other

• Add eval ci by @RunningLeon in #1060
• Ete testcase add more models by @zhulinJulia24 in #1077
• Fix win ci by @irexyc in #1132
• bump version to v0.2.4 by @lvhan028 in #1171

Full Changelog: v0.2.3...v0.2.4

LMDeploy Release V0.2.3

What's Changed

🚀 Features

• Support loading model from modelscope by @irexyc in #1069

💥 Improvements

• Remove caching tokenizer.json by @grimoire in #1074
• Refactor get_logger to remove the dependency of MMLogger from mmengine by @yinfan98 in #1064
• Use TM_LOG_LEVEL environment variable first by @zhyncs in #1071
• Speed up the initialization of w8a8 model for torch engine by @yinfan98 in #1088
• Make logging.logger's behavior consistent with MMLogger by @irexyc in #1092
• Remove owned_session for torch engine by @grimoire in #1097
• Unify engine initialization in pipeline by @irexyc in #1085
• Add skip_special_tokens in GenerationConfig by @grimoire in #1091
• Use default stop words for turbomind backend in pipeline by @irexyc in #1119
• Add input_token_len to Response and update Response document by @AllentDan in #1115

🐞 Bug fixes

• Fix fast tokenizer swallows prefix space when there are too many white spaces by @AllentDan in #992
• Fix turbomind CUDA runtime error invalid argument by @zhyncs in #1100
• Add safety check for incremental decode by @AllentDan in #1094
• Fix device type of get_ppl for turbomind by @RunningLeon in #1093
• Fix pipeline init turbomind from workspace by @irexyc in #1126
• Add dependency version check and fix ignore_eos logic by @grimoire in #1099
• Change configuration_internlm.py to configuration_internlm2.py by @HIT-cwh in #1129

📚 Documentations

• Update contribution guide by @zhyncs in #1120

🌐 Other

• Bump version to v0.2.3 by @lvhan028 in #1123

New Contributors

• @yinfan98 made their first contribution in #1064

Full Changelog: v0.2.2...v0.2.3

LMDeploy Release V0.2.2

Highlight

English version

• The allocation strategy for k/v cache is changed. The parameter cache_max_entry_count defaults to 0.8. It means the proportion of GPU FREE memory rather than TOTAL memory. The default value is updated to 0.8. It can help prevent OOM issues.
• The pipeline API supports streaming inference. You may give it a try!

from lmdeploy import pipeline
pipe = pipeline('internlm/internlm2-chat-7b')
for item in pipe.stream_infer('hi, please intro yourself'):
    print(item)

• Add api key and ssl to api_server

Chinese version

• TurboMind engine 修改了GPU memory分配策略。k/v cache 内存比例参数 cache_max_entry_count 缺省值变更为 0.8。它表示 GPU空闲内存的比例，不再是 GPU 总内存的比例。
• Pipeline 支持流式输出接口。可以尝试下如下代码：

from lmdeploy import pipeline
pipe = pipeline('internlm/internlm2-chat-7b')
for item in pipe.stream_infer('hi, please intro yourself'):
    print(item)

• api_server 在接口中增加了 api_key

What's Changed

🚀 Features

• add alignment tools by @grimoire in #1004
• support min_length for turbomind backend by @irexyc in #961
• Add stream mode function to pipeline by @AllentDan in #974
• [Feature] Add api key and ssl to http server by @AllentDan in #1048

💥 Improvements

• hide stop-words in output text by @grimoire in #991
• optimize sleep by @grimoire in #1034
• set example values to /v1/chat/completions in swagger UI by @AllentDan in #984
• Update adapters cli argument by @RunningLeon in #1039
• Fix turbomind end session bug. Add huggingface demo document by @AllentDan in #1017
• Support linking the custom built mpi by @lvhan028 in #1025
• sync mem size for tp by @lzhangzz in #1053
• Remove model name when loading hf model by @irexyc in #1022
• support internlm2-1_8b by @lvhan028 in #1073
• Update chat template for internlm2 base model by @lvhan028 in #1079

🐞 Bug fixes

• fix TorchEngine stuck when benchmarking with tp>1 by @grimoire in #942
• fix module mapping error of baichuan model by @grimoire in #977
• fix import error for triton server by @RunningLeon in #985
• fix qwen-vl example by @irexyc in #996
• fix missing init file in modules by @RunningLeon in #1013
• fix tp mem usage by @grimoire in #987
• update indexes_containing_token function by @AllentDan in #1050
• fix flash kernel on sm 70 by @grimoire in #1027
• Fix baichuan2 lora by @grimoire in #1042
• Fix modelconfig in pytorch engine, support YI. by @grimoire in #1052
• Fix repetition penalty for long context by @irexyc in #1037
• [Fix] Support QLinear in rowwise_parallelize_linear_fn and colwise_parallelize_linear_fn by @HIT-cwh in #1072

📚 Documentations

• add docs for evaluation with opencompass by @RunningLeon in #995
• update docs for kvint8 by @RunningLeon in #1026
• [doc] Introduce project OpenAOE by @JiaYingLii in #1049
• update pipeline guide and FAQ about OOM by @lvhan028 in #1051
• docs update cache_max_entry_count for turbomind config by @zhyncs in #1067

🌐 Other

• update ut ci to new server node by @RunningLeon in #1024
• Ete testcase update by @zhulinJulia24 in #1023
• fix OOM in BlockManager by @zhyncs in #973
• fix use engine_config.tp when tp is None by @zhyncs in #1057
• Fix serve api by moving logger inside process for turbomind by @AllentDan in #1061
• bump version to v0.2.2 by @lvhan028 in #1076

New Contributors

• @zhyncs made their first contribution in #973
• @JiaYingLii made their first contribution in #1049

Full Changelog: v0.2.1...v0.2.2

LMDeploy Release V0.2.1

What's Changed

💥 Improvements

• [Fix] interlm2 chat format by @Harold-lkk in #1002

🐞 Bug fixes

• fix baichuan2 conversion by @AllentDan in #972
• [Fix] interlm messages2prompt by @Harold-lkk in #1003

📚 Documentations

• add guide about installation on cuda 12+ platform by @lvhan028 in #988

🌐 Other

• bump version to v0.2.1 by @lvhan028 in #1005

Full Changelog: v0.2.0...v0.2.1

LMDeploy Release V0.2.0

What's Changed

🚀 Features

• Support internlm2 by @lvhan028 in #963
• [Feature] Add params config for api server web_ui by @amulil in #735
• [Feature]Merge lmdeploy lite calibrate and lmdeploy lite auto_awq by @pppppM in #849
• Compute cross entropy loss given a list of input tokens by @lvhan028 in #830
• Support QoS in api_server by @sallyjunjun in #877
• Refactor torch inference engine by @lvhan028 in #871
• add image chat demo by @irexyc in #874
• check-in generation config by @lvhan028 in #902
• check-in ModelConfig by @AllentDan in #907
• pytorch engine config by @grimoire in #908
• Check-in turbomind engine config by @irexyc in #909
• S-LoRA support by @grimoire in #894
• add init in adapters by @grimoire in #923
• Refactor LLM inference pipeline API by @AllentDan in #916
• Refactor gradio and api_server by @AllentDan in #918
• Add request distributor server by @AllentDan in #903
• Upgrade lmdeploy cli by @RunningLeon in #922

💥 Improvements

• add top_k value for /v1/completions and update the documents by @AllentDan in #870
• export "num_tokens_per_iter", "max_prefill_iters" and etc when converting a model by @lvhan028 in #845
• Move api_server dependencies from serve.txt to runtime.txt by @lvhan028 in #879
• Refactor benchmark bash script by @lvhan028 in #884
• Add test case for function regression by @zhulinJulia24 in #844
• Update test triton CI by @RunningLeon in #893
• Update dockerfile by @RunningLeon in #891
• Perform fuzzy matching on chat template according to model path by @AllentDan in #839
• support accessing lmdeploy version by lmdeploy.version_info by @lvhan028 in #910
• Remove flash-attn dependency of lmdeploy lite module by @lvhan028 in #917
• Improve setup by removing pycuda dependency and adding cuda runtime and cublas to RPATH by @irexyc in #912
• remove unused settings in turbomind engine config by @irexyc in #921
• Cleanup fixed attributes in turbomind engine config by @irexyc in #928
• fix get_gpu_mem by @grimoire in #934
• remove instance_num argument by @AllentDan in #931
• Fix matching results of several chat templates like llama2, solar, yi and so on by @AllentDan in #925
• add pytorch random sampling by @grimoire in #930
• suppress turbomind chat warning by @irexyc in #937
• modify type hint of api to avoid import _turbomind by @AllentDan in #936
• accelerate pytorch benchmark by @grimoire in #946
• Remove tp from pipline argument list by @lvhan028 in #947
• set gradio default value the same as chat.py by @AllentDan in #949
• print help for cli in case of failure by @RunningLeon in #955
• return dataclass for pipeline by @AllentDan in #952
• set random seed when it is None by @AllentDan in #958
• avoid run get_logger when import lmdeploy by @RunningLeon in #956
• support mlp s-lora by @grimoire in #957
• skip resume logic for pytorch backend by @AllentDan in #968
• Add ci for ut by @RunningLeon in #966

🐞 Bug fixes

• add tritonclient req by @RunningLeon in #872
• Fix uninitialized parameter by @lvhan028 in #875
• Fix overflow by @irexyc in #897
• Fix data offset by @AllentDan in #900
• Fix context decoding stuck issue when tp > 1 by @irexyc in #904
• [Fix] set scaling_factor 1 forcefully when sequence length is less than max_pos_emb by @lvhan028 in #911
• fix pytorch llama2 with new transformers by @grimoire in #914
• fix local variable 'output_ids' referenced before assignment by @irexyc in #919
• fix pipeline stop_words type error by @AllentDan in #929
• pass stop words to openai api by @AllentDan in #887
• fix profile generation multiprocessing error by @AllentDan in #933
• Miss init.py in modeling folder by @lvhan028 in #951
• fix cli with special arg names by @RunningLeon in #959
• fix logger in tokenizer by @RunningLeon in #960

📚 Documentations

• Improve user guide by @lvhan028 in #899
• Add user guide about pytorch engine by @grimoire in #915
• Update supported models and add quick start section in README by @lvhan028 in #926
• Fix scripts in benchmark doc by @panli889 in #941
• Update get_started and w4a16 tutorials by @lvhan028 in #945
• Add more docstring to api_server and proxy_server by @AllentDan in #965
• stable api_server benchmark result by a non-zero await by @AllentDan in #885
• fix pytorch backend can not properly stop by @AllentDan in #962
• [Fix] Fix calibrate bug when transformers>4.36 by @pppppM in #967

🌐 Other

• bump version to v0.2.0 by @lvhan028 in #969

New Contributors

• @amulil made their first contribution in #735
• @zhulinJulia24 made their first contribution in #844
• @sallyjunjun made their first contribution in #877
• @panli889 made their first contribution in #941

Full Changelog: v0.1.0...v0.2.0

LMDeploy Release V0.1.0

What's Changed

🚀 Features

• Add extra_requires to reduce dependencies by @RunningLeon in #580
• TurboMind 2 by @lzhangzz in #590
• Support loading hf model directly by @irexyc in #685
• convert model with hf repo_id by @irexyc in #774
• Support turbomind bf16 by @grimoire in #803
• support image_embs input by @irexyc in #799
• Add api.py by @AllentDan in #805

💥 Improvements

• Fix Tokenizer encode by @AllentDan in #645
• Optimize for throughput by @lzhangzz in #701
• Replace mmengine with mmengine-lite by @zhouzaida in #715
• Set the default value of max_context_token_num 1 by @lvhan028 in #761
• add triton server test and workflow yml by @RunningLeon in #760
• improvement(build): enable ninja and gold linker by @tpoisonooo in #767
• Report first-token-latency and token-latency percentiles by @lvhan028 in #736
• Unify prefill & decode passes by @lzhangzz in #775
• add cuda12.1 build check ci by @irexyc in #782
• auto upload cuda12.1 python pkg to release when create new tag by @irexyc in #784
• Report the inference benchmark of models with different size by @lvhan028 in #794
• Simplify block manager by @lzhangzz in #812
• Disable attention mask when it is not needed by @lzhangzz in #813
• FIFO pipe strategy for api_server by @AllentDan in #795
• simplify the header of the benchmark table by @lvhan028 in #820
• add encode for opencompass by @AllentDan in #828
• fix: awq should save bin files by @hscspring in #793
• Support building docker image manually in CI by @RunningLeon in #825

🐞 Bug fixes

• Fix init of batch state by @lzhangzz in #682
• fix turbomind stream canceling by @grimoire in #686
• [Fix] Fix load_checkpoint_in_model bug by @HIT-cwh in #690
• Fix wrong eos_id and bos_id obtained through grpc api by @lvhan028 in #644
• Fix cache/output length calculation by @lzhangzz in #738
• [Fix] Skip empty batch by @lzhangzz in #747
• [Fix] build docker image failed since packaging is missing by @lvhan028 in #753
• [Fix] Rollback the data type of input_ids to TYPE_UINT32 in preprocessor's proto by @lvhan028 in #758
• fix turbomind build on sm<80 by @grimoire in #754
• Fix early-exit condition in attention kernel by @lzhangzz in #788
• Fix missed arguments when benchmark static inference performance by @lvhan028 in #787
• fix extra colon in InternLMChat7B template by @C1rN09 in #796
• Fix local kv head num by @lvhan028 in #806
• Fix out-of-bound access by @lzhangzz in #809
• Set smem size for repetition penalty kernel by @lzhangzz in #818
• Fix cache verification by @lzhangzz in #821
• fix finish_reason by @AllentDan in #816
• fix turbomind awq by @grimoire in #847
• Fix stop requests by await before turbomind queue.get() by @AllentDan in #850
• [Fix] Fix meta tensor error by @pppppM in #848
• Fix cuda reinitialization in a multiprocessing setting by @grimoire in #862
• launch gradio server directly with hf model by @AllentDan in #856
• fix typo by @grimoire in #769
• Add chat template for Yi by @AllentDan in #779
• fix api_server stop_session and end_session by @AllentDan in #835
• Return the iterator after erasing it from a map by @irexyc in #864

📚 Documentations

• [Docs] Update Supported Matrix by @pppppM in #679
• [Docs] Update KV8 Docs by @pppppM in #681
• [Doc] Update restful api doc by @AllentDan in #662
• Check-in user guide about turbomind config by @lvhan028 in #680
• Update benchmark user guide by @lvhan028 in #763
• [Docs] Fix typo in restful_api user guide by @maxchiron in #858
• [Docs] Fix typo in restful_api user guide by @maxchiron in #859

🌐 Other

• bump version to v0.1.0a0 by @lvhan028 in #709
• bump version to 0.1.0a1 by @lvhan028 in #776
• bump version to v0.1.0a2 by @lvhan028 in #807
• bump version to v0.1.0 by @lvhan028 in #834

New Contributors

• @zhouzaida made their first contribution in #715
• @C1rN09 made their first contribution in #796
• @maxchiron made their first contribution in #858

Full Changelog: v0.0.14...v0.1.0

LMDeploy Release V0.1.0a2

What's Changed

💥 Improvements

• Unify prefill & decode passes by @lzhangzz in #775
• add cuda12.1 build check ci by @irexyc in #782
• auto upload cuda12.1 python pkg to release when create new tag by @irexyc in #784
• Report the inference benchmark of models with different size by @lvhan028 in #794
• Add chat template for Yi by @AllentDan in #779

🐞 Bug fixes

• Fix early-exit condition in attention kernel by @lzhangzz in #788
• Fix missed arguments when benchmark static inference performance by @lvhan028 in #787
• fix extra colon in InternLMChat7B template by @C1rN09 in #796
• Fix local kv head num by @lvhan028 in #806

📚 Documentations

• Update benchmark user guide by @lvhan028 in #763

🌐 Other

• bump version to v0.1.0a2 by @lvhan028 in #807

New Contributors

• @C1rN09 made their first contribution in #796

Full Changelog: v0.1.0a1...v0.1.0a2