Commit graph

16 commits

Author SHA1 Message Date
Limour
f165048a69
feat: add support for KV cache quantization options (#1307)
* add KV cache quantization options

https://github.com/abetlen/llama-cpp-python/discussions/1220
https://github.com/abetlen/llama-cpp-python/issues/1305

* Add ggml_type

* Use ggml_type instead of string for quantization

* Add server support

---------

Co-authored-by: Andrei Betlen <abetlen@gmail.com>
2024-04-01 10:19:28 -04:00
Andrei Betlen
727d60c28a misc: Format 2024-02-28 14:27:40 -05:00
Andrei Betlen
0d37ce52b1 feat: Update llama.cpp 2024-02-28 14:27:16 -05:00
Andrei
4d574bd765
feat(server): Add support for pulling models from Huggingface Hub (#1222)
* Basic support for hf pull on server

* Add hf_model_repo_id setting

* Update README
2024-02-26 14:35:08 -05:00
Andrei Betlen
dcf38f6141 fix: remove prematurely commited change 2024-02-25 21:00:37 -05:00
Andrei Betlen
2292af5796 feat: Update llama.cpp 2024-02-25 16:53:58 -05:00
Andrei Betlen
fdce078cb9 feat: Update llama.cpp 2024-02-17 00:37:51 -05:00
Andrei
fb762a6041
Add speculative decoding (#1120)
* Add draft model param to llama class, implement basic prompt lookup decoding draft model

* Use samplingcontext for sampling

* Use 1d array

* Use draft model for sampling

* Fix dumb mistake

* Allow for later extensions to the LlamaDraftModel api

* Cleanup

* Adaptive candidate prediction

* Update implementation to match hf transformers

* Tuning

* Fix bug where last token was not used for ngram prediction

* Remove heuristic for num_pred_tokens (no benefit)

* fix: n_candidates bug.

* Add draft_model_num_pred_tokens server setting

* Cleanup

* Update README
2024-01-31 14:08:14 -05:00
Andrei
da003d8768
Automatically set chat format from gguf (#1110)
* Use jinja formatter to load chat format from gguf

* Fix off-by-one error in metadata loader

* Implement chat format auto-detection
2024-01-29 14:22:23 -05:00
Andrei Betlen
b8fc1c7d83 feat: Add ability to load chat format from huggingface autotokenizer or tokenizer_config.json files. 2024-01-18 21:21:37 -05:00
Andrei Betlen
48c3b77e6f Offload KQV by default 2024-01-18 11:08:57 -05:00
Kyle Mistele
9c36688b33
fix(cli): allow passing n_ctx=0 to openAI API server args to use model n_ctx_train field per #1015 (#1093) 2024-01-16 18:54:06 -05:00
Andrei Betlen
84615adbc6 Add split_mode option. Closes #1085 2024-01-15 12:49:20 -05:00
Phil H
76aafa6149
Implement GGUF metadata KV overrides (#1011)
* Implement GGUF metadata overrides

* whitespace fix

* Fix kv overrides.

* Fix pointer and pickle

* Match llama.cpp kv_overrides cli argument

---------

Co-authored-by: Andrei <abetlen@gmail.com>
2024-01-15 12:29:29 -05:00
Andrei Betlen
522aecb868 docs: add server config docs 2023-12-22 14:37:24 -05:00
Dave
12b7f2f4e9
[Feat] Multi model support (#931)
* Update Llama class to handle chat_format & caching

* Add settings.py

* Add util.py & update __main__.py

* multimodel

* update settings.py

* cleanup

* delete util.py

* Fix /v1/models endpoint

* MultiLlama now iterable, app check-alive on "/"

* instant model init if file is given

* backward compability

* revert model param mandatory

* fix error

* handle individual model config json

* refactor

* revert chathandler/clip_model changes

* handle chat_handler in MulitLlama()

* split settings into server/llama

* reduce global vars

* Update LlamaProxy to handle config files

* Add free method to LlamaProxy

* update arg parsers & install server alias

* refactor cache settings

* change server executable name

* better var name

* whitespace

* Revert "whitespace"

This reverts commit bc5cf51c64a95bfc9926e1bc58166059711a1cd8.

* remove exe_name

* Fix merge bugs

* Fix type annotations

* Fix type annotations

* Fix uvicorn app factory

* Fix settings

* Refactor server

* Remove formatting fix

* Format

* Use default model if not found in model settings

* Fix

* Cleanup

* Fix

* Fix

* Remove unnused CommandLineSettings

* Cleanup

* Support default name for copilot-codex models

---------

Co-authored-by: Andrei Betlen <abetlen@gmail.com>
2023-12-22 05:51:25 -05:00