fb5a5971d3
feat: proxy server address mapping to the model server ( #461 )
...
* feat: proxy server address mapping to the model server
* fix: add swagger in Config
* refactor: add_proxy_server
* fix: missing semicolo
2023-09-21 07:06:51 +00:00
de3a2271d6
fix(tabby): fix swagger's local server use local port ( #458 )
...
* fixed: swagger's local server use local port
* fix: extract fn add_localhost_server
* fix: add_localhost_server return doc
2023-09-19 04:36:08 +00:00
c107c991ff
chore: bump tabby version to 0.1.1
2023-09-17 17:09:56 +08:00
97eeb6b926
feat: update llama.cpp to fetch latest starcoder support ( #452 )
...
* feat: bump llama.cpp to HEAD
* fix: turn off add_bos by default
2023-09-16 03:41:49 +00:00
86e48afbe0
feat: bump ctranslate2 to HEAD
2023-09-16 02:40:23 +08:00
30afa19bc0
feat: add LLAMA_CPP_LOG_LEVEL to control log level of llama.cpp ( #436 )
2023-09-12 14:41:39 +00:00
1e7ecce697
fix: usize overflow issue in ctranslate2 max_length truncation
2023-09-12 20:56:35 +08:00
1ccf9b2323
refactor: run make fix
2023-09-11 12:58:38 +08:00
09efa1b22b
docs: add client extensions link in swagger landing page
2023-09-11 12:55:42 +08:00
e3c4a77fff
feat: add support fastchat http bindings ( #421 )
...
* feat: add support fastchat http bindings
Signed-off-by: Lei Wen <wenlei03@qiyi.com>
Co-authored-by: Lei Wen <wenlei03@qiyi.com>
2023-09-10 22:17:58 +08:00
f0ed366420
feat: add support vertex-ai http bindings ( #419 )
...
* feat: add support vertex-ai http bindings
* support prefix / suffix
2023-09-09 11:22:58 +00:00
17397c8c8c
feat: add http api bindings ( #410 )
...
* feat: add http-api-bindings
* feat: add http-api-bindings
* hand max_input_length
* rename
* update
* update
* add examples/simple.rs
* update
* add default value for stop words
* update
* fix lint
* update
2023-09-09 03:59:42 +00:00
ad3b974d5c
feat: implement input truncation for llama-cpp-bindings ( #416 )
...
* feat: implement input truncation for llama-cpp-bindings
* set max input length to 1024
* fix: batching tokens with n_batches
* fix batching
2023-09-09 00:20:51 +08:00
87b6b34120
feat: implement input truncation with options.max_input_length ( #415 )
2023-09-08 10:01:03 +00:00
e780031ed6
feat: add ggml fp16 / q8_0 files ( #407 )
...
* feat: add ggml fp16 / q8_0 files
* add q8_0.gguf to optional download files
* add download options to split ctranslate2 files and ggml files
2023-09-06 17:12:29 +00:00
a207520571
feat: turn on metal device by default on macosx / aarch64 devices ( #398 )
2023-09-05 13:03:49 +08:00
d85cd81139
fix: ensure default suffix to be non-empty ( #400 )
2023-09-05 03:45:29 +00:00
e93a971d0e
feat: tune llama metal backend performance ( #393 )
...
* feat: support eos based stop
* feat: print performance stats after each inference
* update llama.cpp
* update commits
2023-09-05 10:14:29 +08:00
2472cf3b55
test: use function call style snippet for prompt builder unit test ( #395 )
...
* test: better tests for build_prefix()
* chore
* chore: resolve comments
2023-09-04 04:54:18 +00:00
74073aa77a
test: add build prefix test and debug chars counting [TAB-184] ( #394 )
...
* test: add count char test
* chore: fix lint
* chore
* chore
2023-09-03 20:57:26 +08:00
3c7c8d9293
feat: add cargo test to github actions and run only unit tests in ci [TAB-185] ( #390 )
...
* feat: add cargo test to github actions
* chore: fix lint
* chore: add openblas dependency
* chore: update build dependency
* chore: resolve comments
* chore: fix lint
* chore: fix lint
* chore: test installing dependencies
* chore: refactor integ test
* update ci
* cleanup
---------
Co-authored-by: Meng Zhang <meng@tabbyml.com>
2023-09-03 05:04:52 +00:00
c8339a2912
refactor: use TabbyML/llama.cpp submodule
2023-09-03 12:38:54 +08:00
3acd5d9bc4
refactor: remove llama.cpp subtree
2023-09-03 12:37:26 +08:00
92c8ae8ee7
feat: embed ggml-metal.metal
2023-09-03 10:41:03 +08:00
ed6c5b2e60
Merge commit 'aad80a58b07836bfbf6aedd50993bc54b4257388' as 'crates/llama-cpp-bindings/llama.cpp'
2023-09-03 10:07:10 +08:00
d4137463ef
remove llama.cpp submodule
2023-09-03 10:04:26 +08:00
e360b438b4
fix lint
2023-09-03 10:01:28 +08:00
3f7aa99b0d
feat: support cancellation in llama backend
2023-09-03 09:59:40 +08:00
3573d4378e
feat: llama.cpp for metal support [TAB-146] ( #391 )
...
* feat: init commit adding llama-cpp-bindings
* add llama.cpp submodule
* add LlamaEngine to hold llama context / llama model
* add cxxbridge
* add basic greedy sampling
* move files
* make compile success
* connect TextGeneration with LlamaEngine
* experimental support llama.cpp
* add metal device
* add Accelerate
* fix namespace for llama-cpp-bindings
* fix lint
* move stepping logic to rust
* add stop words package
* use stop-words in ctranslate2-bindings
* use raw string for regex
* use Arc<Tokenizer> for sharing tokenizers
* refactor: remove useless stop_words_encoding_offset
* switch to tokenizers 0.13.4-rc.3
* fix lints in cpp
* simplify implementation of greedy decoding
* feat: split metal feature for llama backend
* add ci
* update ci
* build tabby bin in ci build
2023-09-03 09:59:07 +08:00
5dff349801
add single line comment to languages so they can be used in prompting [TAB-181] ( #388 )
...
* chore: add comment signs to extended languages
* Update crates/tabby/src/serve/completions/prompt.rs
---------
Co-authored-by: Meng Zhang <meng@tabbyml.com>
2023-09-01 03:43:27 +00:00
63c00494f3
test: unit tests to prompt builder [TAB-180] ( #387 )
...
* test: unit tests to prompt builder
* chore: fix typo
* chore: fix lint
* chore: resolve comments
2023-09-01 09:20:20 +08:00
90aadad3ce
feat: map js,ts,jsx and tsx to js-ts as unified language [TAB-181] ( #386 )
...
* feat: reduce js, ts, jsx and tsx to js-ts
* chore: refactor and add language reducing to both indexing and dataset jobs
* chore: only reduce language in dataset job
* chore: only reduce language in index job
* chore: fix lint
* chore: resolve comments
2023-08-31 17:21:39 +00:00
e5598e63f2
feat: extend language [TAB-181] ( #385 )
...
* feat: extend indexer's language support
* feat: extend language support
* chore: add support for mjs and mts
* chore: fix lint
2023-08-31 07:36:57 +00:00
c44a9c7195
fix: correct git_describe in /health ( #383 )
...
* fix: add missing Version component in OpenAPI definition
* fix: allow tag / dirty in git describe
2023-08-31 01:06:36 +00:00
57baecb370
fix: switch default running backend to openblas on x86 linux ( #380 )
2023-08-30 14:19:35 +00:00
054aefaf15
chore: add linux static build ( #379 )
...
* chore: add linux static build
* add touch
* update build env
* add sudo
* fix: protobuf ubuntu target
2023-08-30 18:45:05 +08:00
3526ca3164
chore: build with ruy (cpu only) on static mode for linux. ( #378 )
...
* chore: build with ruy (cpu only) on static mode for linux.
* update cmake min version
2023-08-30 18:04:40 +08:00
fc9a623e72
feat: add logging on server starting ( #372 )
2023-08-28 06:12:00 +00:00
65836ee199
feat: add stop words encoding offset for ctranslate model config ( #371 )
...
* feat: add stop words encoding offset for ctranslate model config
* feat: set default suffix to \n
* add special treatment for bytefallback tokens
2023-08-28 14:07:01 +08:00
2a91a21787
feat: add gpu info to health state [TAB-162] ( #364 )
...
* feat: add gpu info to health response
* chore: error handling
* chore: refactor cpu manager code
* chore: typo
* chore: fix context mutability
* chore: fix context mutability
* feat: add link to NVML lib
* chore: refactor
* lint
* chore: resolve comments
* chore: fix typo
* chore: fix
* chore: resolve comments
* chore: fix
* chore: resolve comments
2023-08-21 18:06:38 +08:00
b1ad936033
feat: add version information in health state. ( #363 )
...
* feat: add git_hash in health state
* add more version information in health state
2023-08-20 15:21:12 +00:00
df45573501
feat: reduce ServeHealth event to every 300s to reduce event volume ( #362 )
2023-08-20 12:36:59 +00:00
2026b4dd0e
feat: add architecture/cpu info to health api response [TAB-162] ( #355 )
...
* feat: add architecture, cpu and gpu info to health command
* chore: fix
* chore: fix
* chore: fix
* chore: fix lint
* chore: fix lint
* chore: remove gpu
* chore: resolve comments
* chore: resolve comments
* Update health.rs
---------
Co-authored-by: Meng Zhang <meng@tabbyml.com>
2023-08-15 15:22:03 +00:00
dbc89831b1
feat: add serve health heartbeat ( #343 )
...
* add serve health tracking
* fix lint
* fix
2023-08-09 08:08:42 +00:00
d0f6ad2d2a
feat: add anonymous usage tracker ( #342 )
...
* feat: add anonymous usage tracker
* improve deps
* update
* update
2023-08-09 07:31:13 +00:00
220fcc0d65
fix: make `config.experimental` optional ( #339 )
...
* fix: make config.experimental` optional
* add unit test for empty toml config
2023-08-07 09:53:00 +00:00
4eaae27ed3
Update Cargo.toml ( #331 )
2023-08-03 19:55:00 +08:00
6a50902ca7
fix: support ctranslate2 rev7 vocab files (.json) ( #327 )
2023-08-02 13:36:31 +00:00
57c811b30f
fix: improve download logging ( #325 )
...
* Suggest use `-it` so docker run generate progress bar of downloading information properly
* add info! log for model download
2023-08-02 06:30:35 +00:00
b8308b7118
refactor: extract TextGeneration trait ( #324 )
...
* add tabby-inference
* extract TextGeneration trait
* format
* Rename TextInferenceEngine to CTranslate2Engine
2023-08-02 06:12:51 +00:00
83e1cf76d8
feat: Upgrade ctranslate2 to v3.17.1 ( #323 )
2023-08-02 05:46:08 +00:00
95bd53ac9c
feat: add select kind param. Supported editors could log line select … ( #299 )
...
* feat: add select kind param. Supported editors could log line select or block select
* fix lint
2023-07-16 16:02:40 +08:00
be5fe0d737
feat: add rust prompt rewrite support ( #296 )
2023-07-13 09:31:44 +00:00
4388fd0050
feat: support prompt rewriting ( #295 )
...
* refactor: extract PromptBuilder
* feat: load tantivy index in prompt builder
* integrate with searcher
* add enable_prompt_rewrite to control rewrite behavior
* nit docs
* limit 1 snippet per identifier
* extract magic numbers
2023-07-13 09:05:41 +00:00
19586a4926
fix: fix bad indexing field of tag indexer ( #293 )
2023-07-13 07:19:46 +00:00
24d17dc1ed
feat: switching index to tag based ( #265 )
...
* add tag based index
* update for python
* fix lint
2023-07-13 06:25:39 +00:00
be90047477
fix: fix int8 compute type, fix auto compute type selection (include float32 into consideration for cuda compute capability <= 6.0) ( #291 )
2023-07-12 11:09:38 +08:00
e822d1857f
fix: add retry strategy in tabby download ( #285 )
2023-07-09 05:18:19 +00:00
9ca1f7e5f1
fix: add additional whitespace to match tokens that combining space and li… ( #270 )
...
* fix: add additional whitespace to match tokens that combining space and line break
* fix lint
2023-06-25 01:15:52 +00:00
631cff3aed
docs: update url of playground server
2023-06-23 18:55:23 -07:00
8879ed7655
feat: add experimental self host setup ( #261 )
2023-06-23 12:13:22 -07:00
59b03ae477
fix: use full clone in scheduler
2023-06-22 15:06:44 -07:00
fcbc5edc55
Revert "feat: add /experimental/search endpoint ( #258 )" ( #260 )
...
This reverts commit 04980160e5
2023-06-22 14:23:35 -07:00
04980160e5
feat: add /experimental/search endpoint ( #258 )
...
* feat: add /experimental/search endpoint
* fix format
2023-06-22 20:47:32 +00:00
8c1f6c2806
feat(scheduler): add java / python tags support ( #259 )
2023-06-22 20:25:56 +00:00
7ed5dd584d
feat: experiment ctags support in scheduler ( #207 )
...
* experiment ctags support
* add document.rs
* extract Document to common
* integrate tags into dataset builder
* skip if none
* do not add scheduler in client binary
* fix fmt
2023-06-21 19:48:13 -07:00
6eae16d475
fix: typo in openapi documentation
2023-06-16 20:22:38 -07:00
8ee700089f
feat: do not use fim template when suffix is empty string
2023-06-15 09:27:32 -07:00
9abf1a7521
fix: optional file should be put in cache key to avoid internet access when prefer_local_files = true ( #241 )
...
* feat: when file is 404, cache as NotFound
* explicitly mark optional file
* refactor
2023-06-15 04:13:52 +00:00
d572cf7d6d
api: add user field in completion api
2023-06-14 10:50:03 -07:00
93d5d8b297
docs: update website openapi.json
2023-06-13 13:13:03 -07:00
b2734aed59
feat: returns more information in /v1/health
2023-06-13 13:11:20 -07:00
9c9e46c6f4
feat: support set compute_type through commandline arguments
2023-06-13 12:04:07 -07:00
ba7e04d030
refactor: remove admin
2023-06-13 11:37:55 -07:00
5985d91782
fix: use int8_float16 to fix SantaCoder-1B ( #237 )
...
#236
2023-06-13 01:13:06 -07:00
a8a145d9b3
feat: add metrics in dataset ( #235 )
2023-06-12 19:39:43 +00:00
2983f70d22
feat: build index from dataset ( #234 )
2023-06-12 19:21:27 +00:00
9af6607f85
feat: add dataset building ( #231 )
...
* feat: build dataset in scheduler
* add file rotate
2023-06-12 15:19:43 +00:00
3b7153ba23
docs: update website
2023-06-11 12:28:23 -07:00
de546b03fe
feat: add otlp-endpoint for OpenTelemetry support [TAB-67] ( #227 )
...
* feat: add otlp-endpoint for OpenTelemetry support
* set default log level for axum tracing to INFO
* update build enviornment
* update
2023-06-10 22:46:25 -07:00
6180b32980
feat: add /v1/health ( #226 )
...
* feat: add /v1/health
* fix fmt
2023-06-10 22:37:42 -07:00
6718afbf67
fix: server should still support prompt only use case
2023-06-10 00:39:32 -07:00
eecdde0269
fix: vocabulary.txt / shared_vocabulary.txt should be optional ( #225 )
2023-06-09 06:49:09 +00:00
40460885b0
docs: improve website theming
2023-06-07 11:17:00 -07:00
1aaf29c968
docs: switch openapi docs ( #215 )
...
* update openapi
* update
* fix: shared_vocabulary is not a required file
* docs: improve docs
2023-06-07 01:58:05 -07:00
4cb672ec39
feat: improve error handling and messages [TAB-58] ( #213 )
...
* add fatal macro
* switch expect to fatal
* improve error handling of serve
* improve error handling on download module
* improve error handling in scheduler
* improve error handling
* fmt
* fmt
2023-06-07 02:02:58 +00:00
fd1baff8d5
feat: support stop sequences [TAB-52] ( #212 )
...
* refactor: pass step and string token to callback
* add token to callback
* add stop regexp
* implement stop words logic
* pass token_ids from inference
* improve effiency of regexp match with reversed regex
* fmt
* add typescript and javascript stop words
* add cache for stop words regexp
2023-06-06 23:28:58 +00:00
007a40c582
feat: support early stop [TAB-51] ( #208 )
...
* bump ctranslate2 to v3.15.0
* enable early stop
* support early stop
2023-06-06 12:46:17 +00:00
249d51d0f5
feat: add indexer [TAB-17] ( #199 )
...
* add basic indexer
* formatting
2023-06-05 22:18:10 +00:00
272dde9769
refactor: rust nightly format ( #197 )
...
* chore: turn on group format
* turn on nightly fmt
2023-06-05 14:17:07 -07:00
f4442b104f
docs: usage string for scheduler
2023-06-05 12:57:18 -07:00
e8a33312bb
refactor: extract download into tabby-download ( #195 )
...
* refactor: extract download into tabby-download
* remove unused deps
2023-06-05 18:40:24 +00:00
e8b1c10738
feat: add `tabby scheduler` command ( #194 )
...
* feat: add `tabby scheduler` command
* update test cases
* fix fmt
2023-06-05 18:29:38 +00:00
da02d471a9
chore: add tabby-scheduler [TAB-17] ( #192 )
...
* add scheduler
* update fmt
* add integration tests for scheduler
2023-06-05 03:08:43 +00:00
2bf5bcd0cf
refactor: extract TextInferenceEngineImpl to reduce duplications between EncoderDecoderImpl and DecoderImpl #189
2023-06-04 22:28:39 +00:00
6de61f45bb
chore: mark thread safety [TAB-52] ( #186 )
...
* mark thread safety
* use shared_ptr to ensure thread safety
* fmt
2023-06-04 06:23:31 +00:00
775576b53e
fix: only use prompt_template when suffix presents [TAB-46] ( #184 )
...
* fix: only use prompt_template when suffix presents
* lint
2023-06-03 17:29:04 +00:00
2779da3cba
feat: supports FIM inference [TAB-46] ( #183 )
...
* Add prefix / suffix
* update
* feat: support segments in inference
* chore: add tabby.json in model repository to store prompt_template
* make prompt_template optional.
* download tabby.json in downloader
2023-06-02 16:47:48 -07:00
950a7a795f
fix: when model_id is an local dir, don't try to download model from remote
2023-06-02 13:48:53 -07:00
3cac2607e7
refactor: improve error handlings, fix clippy warnings ( #181 )
...
* refactor: minor improvements on error handling
* refactor: cleanup error handlings
* update
* update
* fix
* add clippy / test workflow
* fix clippy
* fix clippy
* update
2023-06-01 17:23:05 -07:00
ca077a3403
feat: ensure model exist before serving ( #180 )
...
* chore: migrate completion to new metadata format
* feat: ensure model exist before serving
2023-06-01 07:26:21 +00:00
9131567257
chore: migrate completion to new metadata format ( #179 )
2023-06-01 07:08:09 +00:00
e8dbd36663
feat: improve download command - support local cache checking behavior ( #178 )
...
* move download.rs
* add metadata
* support prefer local args
* fix format
* replace errorchain with anyhow
2023-06-01 06:42:04 +00:00
b8d67770ee
fix: cors in /, stop words removal ( #171 )
2023-05-30 17:12:10 -07:00
f3b37b253b
feat: add events logger ( #170 )
...
* add common events
* log events in serve
* formatting
* extract serde_conv
* update
* update
* fix logging
* update
2023-05-30 15:44:29 -07:00
d8cee4adac
chore: extract tabby-common ( #169 )
...
* chore: extract tabby-common
* simplify
2023-05-29 23:39:02 -07:00
8956056120
fix: support local model path ( #165 )
2023-05-29 17:40:16 -07:00
418558c05d
chore: cargo fmt
2023-05-29 17:00:23 -07:00
0d11b0e832
Make language field optional in rust implementation ( #164 )
...
* remove download_model.py as we have tabby serve now
* Make language field optional
2023-05-29 16:58:02 -07:00
48796ecd77
feat: add `tabby download` command ( #157 )
...
* simplify fmt-display
* cleanup
* move tabby-admin to reduce nest
* add model downloader
* get rid of model-type
* improve commands
* fix fmt
2023-05-28 14:36:11 -07:00
80d1dd1ca6
Add admin panel (behand --experimental-admin-panel) ( #156 )
...
* add admin
* add feature:admin_panel
* update
* add dist directory
* update
2023-05-28 01:17:32 -07:00
b8309d98cc
Switch to sccache ( #154 )
...
* fix fmt
* fix
* fix test
* fix clippy
* switch to sc cache
* fix
* update
* update
* update
* fix
* add test
* remove clippy
* update
* disable incremental
* update
* simply
2023-05-27 16:20:17 -07:00
0000312460
Add download_model.py. Adjust ctranslate2 backend model structure ( #153 )
...
* adjust
* update
* update
2023-05-27 14:51:12 -07:00
552711a560
Support causal lm (decoder only model) ( #151 )
...
* support
* support causal lm
2023-05-27 01:26:33 -07:00
72ed30e9ff
Build link shared in docker for ctranslate2 ( #150 )
...
* Build link shared in docker
* update
* update
2023-05-27 00:05:56 -07:00
06cf34a007
support static linking of ctranslate2 ( #148 )
...
* support static linking of ctranslate2
* update
* remove submodule rust-cxx-cmake-bridge
* support alwayslink with whole-archive
* update
* move export_libs
* update docker config
* update ctranslate2
* remove
* update
* update build.rs
* parse external libs
* cleanup
* add cargo fmt
2023-05-26 21:34:31 -07:00
7b10340e67
feat: add --port to serve command
2023-05-26 00:32:11 -07:00
c296b83de9
chore: remove unused lock
2023-05-26 00:06:10 -07:00
8dfe49ec6c
feat: support cuda devices in rust tabby ( #149 )
2023-05-25 23:23:07 -07:00
0acc975618
Support linux ctranslate2 cuda build ( #147 )
...
* Support linux build
* add <memory> to fix build error in linux
* add Dockerfile.tabby
* update
* update
* add rust docker image pipeline
* add docker.rust.yml
2023-05-25 18:18:22 -07:00
80588ddd22
fix: remove wrongly added submodule
2023-05-25 15:08:34 -07:00
a2476af373
add ctranslate2-bindings / tabby rust packages ( #146 )
...
* add ctranslate2-bindings
* add fixme for linux build
* turn off shared lib
* add tabby-cli
2023-05-25 14:05:28 -07:00
胡锋