Skip to content
This repository has been archived by the owner on Sep 18, 2024. It is now read-only.

Latest commit

 

History

History
757 lines (366 loc) · 24.1 KB

CHANGELOG.md

File metadata and controls

757 lines (366 loc) · 24.1 KB
Raw

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased] - 2023-MM-DD

Added

Removed

Changed

Fixed

Docs

[0.8.1] - 2023-07-26

Added

Removed

Changed

Fixed

Docs

  • Add tiny model and citation to Readme and docs. (#763)

  • Fix huggingface link of jina embeddings. (#761)

  • Remove redundant text in jina embedding page. (#762)

[0.8.0] - 2023-07-13

Added

  • Add jina embeddings suit. (#757)

  • Add cos_sim helper to finetuner. (#757)

Removed

Changed

  • Finetuner always install torch and other dependencies. (#757)

Fixed

Docs

  • Add Jina embeddings documentation page. (#759)

[0.7.8] - 2023-06-08

Added

  • Support loading models from Jina's huggingface site. (#751)

  • Add multilingual model for training data generation job. (#750)

Removed

Changed

  • Bump black, flake8 and isort (#747)

  • Increase the default num_relations from 3 to 10 for data synthesis job. (#750)

Fixed

  • Fix create synthesis run not accepting DocumentArray as input type. (#748)

Docs

  • Update data synthesis tutorial include english and multilingual model. (#750)

[0.7.7] - 2023-05-24

Added

  • Add support for data generation jobs. (#715)

Removed

Changed

  • Import Document, DocumentArray from finetuner. (#720)

Fixed

Docs

  • Add documentation on using Document and DocumentArray from docarray v1. (#720)

  • Add notebook on data generation. (#745)

[0.7.6] - 2023-04-18

Added

Removed

Changed

  • Install finetuner from source code instead of using a pip package #719.

Fixed

  • Downgrade docarray version #719.

Docs

[0.7.5] - 2023-04-14

Added

  • Add support for python 3.10. (#704)

  • Add error message when experiment is None when creating a run. (#708)

Removed

Changed

  • Use correct schema and model stub for fine-tuning job rather than generation job. (#711)

Fixed

  • Fix error caused by pods restarting. (#709)

Docs

[0.7.4] - 2023-03-29

Added

  • Support pair-wise score document construction from CSV. (#696)

Removed

Changed

  • Refactor load_finetuning_dataset into CSV handlers. (#696)

  • Unify all models names into name-size-lang format. (#700)

  • Do not download pre-trained weights when user downloads the artifact. (#706)

Fixed

Docs

  • Add query-document score into CSV data preparation. (#697)

  • Add CosineSimilarityLoss into advanced loss section. (#697)

  • Add layer-wise learning rate decay into advanced configurations. (#697)

[0.7.3] - 2023-03-16

Added

  • Add support for batch size scaling. (#691)

  • Add functions to retrieve evaluation metrics and example results. (#687)

Removed

Changed

  • Remove unit test and integration test from CD. (#686)

Fixed

Docs

[0.7.2] - 2023-03-02

Added

  • Add support for learning rate schedulers. (#679)

Removed

Changed

  • Remove duplicated documents when parsing unlabeled CSV files. (#678)

  • The scheduler_step options is now part of scheduler_options. (#679)

Fixed

Docs

  • Update documentation on creating training data. (#678)

  • Add notebook to demonstrate use of ArcFaceLoss. (#680)

  • Add section on GeM pooling to advanced topics. (#684)

[0.7.1] - 2023-02-15

Added

  • Add support for new loss and pooling options to the finetuner.fit method. (#664)

  • Add folder for example CSV files. (#663)

  • Add communication between remote-ci job and the pr that triggered it. (#642)

  • Support continuing training from an artifact of a previous run. (#668)

Removed

Changed

  • Use github token provided by dispatcher when running remote-ci. (#640)

Fixed

  • Use python 3.8 in Github actions. (#659)

  • Add proper CSV file for image-image case. (#667)

  • Fix problems with login function in notebooks by bumping hubble version. (#672)

  • Fix URL construction. (#672)

Docs

  • Add page on loss and pooling to advanced-topics. (#664)

  • Remove ResNet backbone support for clip fine-tuning. (#662)

  • Add efficientnet b7 as a new image to image search backbone. (#662)

  • Fix typos, duplicate paragraphs, and wrong formulations. (#666)

  • Add list of articles to README and docs. (#669)

  • Removed section on GeM pooling from advanced topics. (#676)

[0.7.0] - 2023-01-18

Added

  • Add val_split parameter to fit function. (#624)

  • Add core-ci workflow to remotely run the ci of finetuner-core. (#628)

  • Add support for 3d meshes to build_finetuning_dataset. (#638)

Removed

  • Remove cpu parameter from create_run function. (#631)

  • Remove notebook_login function. (#631)

  • Remove support for python 3.7 (#653)

Changed

  • Adjust Finetuner based on API changes for Jina AI Cloud. (#637)

  • Change default experiment_name from current working dir to default. (#637)

  • Use github token provided by dispatcher when running remote-ci. (#640)

Fixed

  • Correctly infer the type of models created using get_model in the build_encoding_dataset function. (#623)

Docs

  • Add before and after section to the example notebooks. (#622)

  • Align text-to-image notebook with its corresponding markdown file. (#621)

  • Change hint in notebooks to use load_uri_to_blob instead of load_uri_to_image_tensor. (#625)

  • Copyedit README.md, changes to language but not contents. (#635)

  • Add multilingual clip colab to readme. (#620)

  • Add tutorial for mesh-to-mesh search. (#638)

  • Add documentation for PointNet++ model and handling 3D mesh dataset. (#638)

  • Add finetuner namespace to artifact names in the documentation. (#649)

  • Rewrite M-CLIP notebook to use German fashion dataset. (#643)

  • New advanced topics section. (#643)

  • Improve developer reference. (#643)

  • Improve walkthrough sections. (#643)

  • Add hints to escape common to prepare csv training data. (#655)

[0.6.7] - 2022-11-25

Added

  • Allow user to control num_items_per_class to sample to each batch. (#614)

Removed

Changed

  • Update commons and stubs versions. (#618)

Fixed

  • Valid configuration of num_items_per_class. (#618)

Docs

  • Add notebook for multilingual CLIP models. (#611)

  • Improve describe_models with task to better organize list of backbones. (#610)

  • Add documentation on using the evaluation callback for CLIP (multiple models). (#615)

  • Ignore callback module in apidoc. (#614)

[0.6.6] - 2022-11-24

This release was broken and was deleted.

Added

Removed

Changed

Fixed

Docs

[0.6.5] - 2022-11-10

Added

  • Add support for CSV files to the EvaluationCallback. (#608)

  • Add support for CSV files to the fit function. (#592)

  • Add support for lists to the encode function. #598

  • Allow user to publish public artifact. #602

Removed

  • Remove connect function. (#596)

Changed

  • Enhance documentation of login functionalities. (#596)

  • Deprecate notebook_login function with login(interactive=True). (#594)

Fixed

  • Correctly use eval_data in the create_run function (#603)

  • Fix links to functions in the documentation. (#596)

Docs

  • Improve documentation on csv reading and run monitoring section. #601

  • Add documentation for WandBLogger. #600

  • Change datasets and hyperparameters for ResNet experiment. (#599)

  • Use login instead of notebook_login in examples. (#605)

[0.6.4] - 2022-10-27

Added

  • Add build_model function to create zero-shot models. (#584)

  • Use latest Hubble with notebook_login support. (#576)

Removed

Changed

  • Use the run config model from finetuner-stubs to create the run config. (#579)

  • Use device parameter to replace cpu to align with docarray. (#577)

  • Update the open clip model names in the table of the backbones. (#580)

  • Show progress while encode batches of documents. (#586)

  • Change device as an optional parameter when calling get_model. (#586)

Fixed

Docs

  • Fix training data name in totally looks like example. (#576)

  • Embed three tasks as three Google Colab notebooks in documentation. (#583)

  • Unify documentation related to cloud storage as Jina AI Cloud. (#582)

  • Replace hub.jina.ai with cloud.jina.ai. (#587)

[0.6.3] - 2022-10-13

Added

  • Support advanced CLIP fine-tuning with WiSE-FT. (#571)

Removed

Changed

  • Change CLIP fine-tuning example in the documentation. (#569)

Fixed

  • Bump flake8 to 5.0.4. (#568)

Docs

  • Add documentation for callbacks. (#567)

[0.6.2] - 2022-09-29

Added

  • Support inference with torch models. (#560)

Removed

Changed

Fixed

  • Freeze hubble client to 0.17.0. (#556)

Docs

  • Fix template html css. (#556)

[0.6.1] - 2022-09-27

Added

  • Add finetuner_version equal to the stubs version in the create run request. (#552)

Removed

  • Improve display of stream log messages. (#549)

Changed

  • Bump hubble client version. (#546)

Fixed

  • Preserve request headers in redirects to the same domain. (#552)

Docs

  • Improve example and install documentation. (#534)

  • Update finetuner executor version in docs. (#543)

[0.6.0] - 2022-09-09

Added

  • Add get_model and encode method to encode docarray. (#522)

  • Add connect function to package. (#532)

Removed

Changed

  • Incorporate commons and stubs to use shared components. (#522)

  • Improve usability of stream_logs. (#522)

  • Improve describe_models with open-clip models. (#528)

  • Use stream logging in the README example. (#532)

Fixed

  • Print logs before run status is STARTED. (#531)

Docs

  • Add inference session in examples. (#529)

[0.5.2] - 2022-08-31

Added

  • Description of get_model and encode function. (#526)

  • Enable wandb callback. (#494)

  • Support log streaming in finetuner client. (#504)

  • Support optimizer and miner options. #517

Removed

Changed

  • Mark fit as login required. (#494)

  • Improve documentation according to remarks we received. (524)

Fixed

  • Replace the artifact name from dot to dash. (#519)

  • Create client automatically if user is already logged in. (#527)

Docs

  • Fix google analytics Id for docs. (#499)

  • Update sphinx-markdown-table to v0.0.16 to get. this fix (#499)

  • Place install instructions in the documentation more prominent. (#518)

[0.5.1] - 2022-07-15

Added

  • Add artifact id and token interface to improve usability. (#485)

Removed

Changed

  • save_artifact should show progress while downloading. (#483)

  • Give more flexibility on dependency versions. (#483)

  • Bump jina-hubble-sdk to 0.8.1. (#488)

  • Improve integration section in documentation. (#492)

  • Bump docarray to 0.13.31. (#492)

Fixed

  • Use uri to represent image content in documentation creating training data code snippet. (#484)

  • Remove out-dated CLIP-specific documentation. (#491)

[0.5.0] - 2022-06-30

Added

  • Docs 0.4.1 backup. (#462)

  • Add Jina integration section in the docs. (#467)

  • Add CD back with semantic release. (#472)

Removed

Changed

  • Refactor the guide for image to image search. (#458)

  • Refactor the guide for text to image search. (#459)

  • Refactor the default hyper-params and docstring format. (#465)

  • Various updates on style, how-to and templates. (#462)

  • Remove time column from Readme table. (#468)

  • Change release trigger to push to main branch. (#478)

Fixed

  • Use finetuner docs links in docs instead of netlify. (#475)

  • Use twine pypi release. (#480)

  • Fix blocked success-all-tests in CI. (#482)

  • Fix documentation render in the login page. (#482)

0.2.2 - 2022-06-16

Added

Removed

  • Remove path and dotenv as dependencies. (#444)

Changed

  • Change default registry to prod for api and hubble. (#447)

  • Polish the documentation structure and references. (#460)

  • Update README.md with latest developments. (#448)

Fixed

0.2.1 - 2022-06-13

Added

Removed

Changed

Fixed

  • docs: fix link references and missing images. (#439)

  • fix: send another request when redirect detected. (#441)

0.2.0 - 2022-06-09

Added

  • Add default values for finetuner HOST and JINA_HUBBLE_REGISTRY. (#410)

  • Expose arguments cpu and num_workers in finetuner.fit. (#411)

  • Add documentation structure and how it works section. (#412)

  • Support passing callbacks to the run configuration. (#415)

  • Add documentation step by step from install to create training data. (#416)

  • Add support for EvaluationCallback. (#422)

  • Docs add developer reference, Jina ecosystem and style fix. (#423)

  • Add support for MLP model. (#428)

  • Add method list_models that returns the available model names. (#428)

  • Organize supported model to model stubs under finetuner.models. (#428)

  • Add a guide for image-to-image retrieval. (#430)

  • Add a guide for text to image fine-tuning with CLIP. (#433)

  • Add template for guides in docs. (#437)

  • Add text to text with Bert guide to docs. (#426)

Removed

Changed

  • Bump docarray to v0.13.17. (#411)

  • Guide user to choose models in list_models. (#419)

  • Expose run methods. (#425)

  • Rename list_models to describe_models. (#428)

  • Rename finetuner.callbacks to finetuner.callback to avoid name collision in __init__.py. (#428)

Fixed

  • Enable saving models for clip without overwriting. (#432)

  • Fix flaky integration test. (#436)

[0.1.0] - 2022-05-23

Added

  • Setup the project structure. (#385)

  • Create experiment endpoints. (#386)

  • Create run endpoints. (#387)

  • Add Hubble authentication. (#388)

  • Add docs and netlify deployment. (#392)

  • Implement Run, Experiment and Finetuner classes on top of the base client. (#391)

  • Basic error handling. (#394)

  • Create a complete version of the run config. (#395)

  • Improve unit testing. (#396)

  • Implement getting run logs. (#400)

  • Add experiment-related methods to finetuner. (#402)

  • Add CD step for PyPI release. (#403)

Removed

  • Delete all unnecessary files from the previous project. (#384)

Changed

  • Change logic behind artifact-id and return jsonified dicts instead of requests.Response objects. (#390)

  • Adapt getting run status. (#400)

Fixed

  • Resolve CI tests. (#398)

  • Fix download_artifact. (#401)