Basaran Versions Save

Basaran is an open-source alternative to the OpenAI text completion API. It provides a compatible streaming API for your Hugging Face Transformers-based text generation models.

v0.7.0

1 year ago

[311a0ce] build(make): bump version to 0.7.0
[74564ab] build(docker): bump version to 0.7.0
[9926724] fix(model): retry should be a temporary workaround
[fdc1e05] feat(playground): add model playground (#9)
[ae53695] fix(model): add a small epsilon to ensure inputs to log are non-zero
[6edcd75] fix(model): remove redundant detach calls
[c9ccc5d] fix(server): data type for top_p should be float

v0.6.0

1 year ago

[3df5a3b] build(make): bump version to 0.6.0
[1882bdd] build(docker): bump version to 0.6.0
[2ae137a] feat(model): add auto retrying
[9842e68] build(deps): add tenacity to dependencies
[203b45c] refactor(model): clean up redundant variables
[1e0a581] build(docker): add default memory fraction
[9cc93e8] feat(model): allow to specify memory fraction
[3ea3cf4] feat(model): use multiple stateful decoders
[79b71fa] feat(decoder): add stream decoder
[fa8e235] build(docker): install dependencies before copying app source
[0d9bc46] refactor(*): update project structure (#6)
[d106b9b] build(docker): use dedicated script for preloading
[f1d3019] build(utils): add script for preloading models
[40610b5] build(deps): remove version hash
[15ebcc0] build(deps): add pytest to dependencies
[71a814f] refactor(model): use encode instead of calling tokenizer
[8156710] build(ci): remove workflow for releasing bundled images

v0.5.0

1 year ago

[303405f] build(make): bump version to 0.5.0
[160d111] build(docker): bump version to 0.5.0
[03d9e8c] feat(model): handle surrogates and whitespace while decoding
[23bf512] build(ci): add workflow for releasing bundled images
[dacfde4] build(make): add target for building bundled images
[88dbb91] docs(contrib): add contribution guide
[87bf8bc] docs(conduct): add code of conduct
[72e7b02] build(docker): shorten max event stream interval
[23b8da3] feat(server): shorten max event stream interval
[fec45fd] build(docker): hide welcome message from bitsandbytes

v0.4.0

1 year ago

[5736c92] build(make): bump version to 0.4.0
[7d020f1] build(docker): bump version to 0.4.0
[0fa6aa6] build(deps): match versions in the base image
[dac0d79] build(docker): use base image provided by nvidia

v0.3.1

1 year ago

[03c04a2] build(make): bump version to 0.3.1
[ae4ccec] build(docker): bump version to 0.3.1
[8da379a] fix(model): inplace update in inference mode

v0.3.0

1 year ago

[acd7bdf] build(make): bump version to 0.3.0
[670c9d4] build(docker): bump version to 0.3.0
[66862cf] feat(model): enable inference mode
[30519fc] build(docker): add bundle examples
[252ea29] feat(server): simplify error handling
[ef39296] build(docker): force stdout and stderr to be unbuffered

v0.2.0

1 year ago

[7b9138e] build(make): bump version to 0.2.0
[0a4aa13] docs(readme): update project description
[ca351a2] build(docker): install torch with cuda 11.6
[eec22cc] build(deps): explicitly specify torch version
[6114c6c] fix(server): convert string to boolean correctly

v0.1.0

1 year ago