- Added tests for CRUD Operations of CO Resource - Deployment [(#431)] (#431)
- Added request-id-prefix header to SDK requests to improve SDK monitoring [(#430)] (#430)
- Added CLI for Model upload [(#429)] (#429)
- Fixed model servicer for Model Upload [(#428)] (#428)
- Added python versions badge to README.md [(#427)] (#427)
- Removed stream tests till stream API is fixed [(#426)] (#426)
- Removed unnecessary prefixes to concept ID added from SDK [(#424)] (#424)
- Upgraded llama-index-core lib version as a security update [(#423)] (#423)
- Added metadata in exported dataset annotations files
- Upgrade to clarifai-grpc 10.9.11
- Improve UX for model upload and fix runners tests [(#420)] (#420)
- Added functionality to Merge Datasets [(#419)] (#419)
- Fix bugs for model upload [(#417)] (#417)
- Fix download_checkpoints and fix run model locally [(#415)] (#415)
- Improve handling missing huggingface_hub package [(#412)] (#412)
- Implement script that allows users to test and run a runner's model locally [(#411)] (#411)
- Improve Model upload experience for cv models [(#408)] (#408)
- Improved the Test Coverage for Dataloaders & Evaluations modules of SDK [(#409)] (#409)
- New streaming predict endpoints(#407)
- New dockerfile for model upload and improvements to upload flow(#406)
- Bug fixes for logger(#405)
- Added CRUD operations for Compute Orchestration resources (Compute cluster, Nodepool, Deployment) [(#402)] (#402)
- Improved logging and fixed issues with downloading checkpoints(#403)
- Refract Model upload and download checkpoints at build time during model upload(#400)
- Added fsspec dependency which would be required in runners for model upload (#398)
- Added MultiModalLoader support (#384)
- Deleted model_serving in this SDK, after the Runners PR has been merged (#391)
- Added validation check in HF loader, if the checkpoints really exit at checkpoint path (#396)
- Remove pydantic dependency from runners in clarifai-python (#395)
- use json logger always in k8s (#393)
- Added a json logger so it's convenient to get logs into logging stacks (#392)
- Added HuggingFaceLoader and added methods in model_upload for download_checkpoints and handling concepts (#390)
- Integrate clarifai-protocol which use to upload model to platform (#389)
- Tests Addition for App, Dataset, Input, Model Classes (#386)
- Upgrade to clarifai-grpc 10.8.7
- Upgrade to clarifai-grpc 10.8.6
- Improved Model Export functionality by adding
Ranges
header (#385)
- Python SDK usage issue on Windows OS due to upgrade in Protobuf library (#380)
- Dataset Annotations bug that returns None if class annotation is not present during export (#382)
- Patch operations for Models and Workflows [(#370)] (#370)
- Addition of Concept Relations Operations [(#371)] (#371)
- Addition of App's Input Count functionality [(#372)] (#372)
- Dataset Annotations bug that returns either class annotation or detection annotation during export (#375)
- Model Export Bug by adding authentication headers (#373)
- Patch operations for Apps and Datasets [(#364)] (#364)
- RAG class to support env variable for
user_id
param (#357) - Search query bug that returns duplicated triplets by removing
PostAnnotationsSearches
and replacing it withPostInputsSearches
(#366) - Search request potentially blocks the users to use different types of filters altogether, fixed it by supporting annotation and input proto filters.(#366)
- Patch operations for input annotations and concepts [(#354)] (#354)
- Getting user id from ENV variables for RAG class (#358)
- Improved rich logging by width addition (#359)
- Dataset export functionality - Added authentication headers to download requests, better exception formatting (#356)
- Moved some convenience features to CLI only to avoid writes to disk (#353)
- Text Features to add random ID as input if input ID is not provided in Dataloader (#351)
- Added BaseClient.from_env() and some new endpoints (#346)
- Upgrade to clarifai-grpc 10.5.0 (#345)
- Upgrade to clarifai-grpc 10.3.4 (#343)
- RAG apps, workflows and other resources automatically setup now use UUIDs in their IDs instead of timestamps to avoid races. (#343)
- Fixed issue with
get_upload_status
overridinglog_warnings
table in log file. (#342) - Use UUIDs in tests to avoid race conditions with timestamps. (#343)
- Hardcoded shcema package to 0.7.5 as it introduced breaking changes. (#343)
- Flag to download model. If export_dir param in
Model().export()
is provided, the exported model will be saved in the specified directory else export status will be shown.(#337) - Label ID support in Dataloaders(label_ids param) and get_proto functions in Inputs class.(#338)
- Logger for
Inputs().upload_annotations
to show full details of failed annotations.(#339)
- RAG upload bug by changing llama-index-core version to 0.10.24 in ImportError message (#336)
- Pagination feature in Search. Added pagination param in
Search()
class and included per_page and page_no params inSearch().query()
(#331) - Alogrithm param in
Search()
(#331)
- Model Upload CLI Doc(#329)
- RAG.setup() bug where if we delete a specific workflow and create another workflow with the same id, by adding timestamp while creating a new prompter model (#332)
RAG.upload()
to support folder of text files.(#332)
- Root certificate support to establish secure gRPC connections by adding
root_certificates_path
param in all the classes and auth helper and updating the grpc to the latest version.(#319) - Missing VERSION and requirements.txt files to setup.py(#320)
- To limit max upload batch size for
Inputs().upload_inputs()
function. Also changed the model version id parameter inconsistency inApp.model()
andModel()
(#317)
- Training status bug by removing constraint of user specifying model_type_id for training_logs and using
load_info()
to get model version details(#321) - Create workflow bug which occured due to the model version id parameter change in #317(#322)
- Unnecessary infra alerts by adding wait time before deleting a model in model training tests (#326)
- Runners from the SDK(#325)
- Dataset version ID support in
app.dataset()
andDataset()
(#315)
- Dataset Export function to internally download the dataset archive zip with the function
Dataset.archive_zip()
(#303) - The backoff iterator to support custom starting count, so different process can have different starting wait times.(#313)
- Removed the key base_embed_model from params.yaml file, since the model training by default considers the base embed model which is set for the app and no need to define it again in params file.(#314)
- File not found error in model serving CLI (#305)
- Workflow YAML schema bug (#308)
- Base URL passing bug (#308)
- Eval Endpoints (#290)
- Eval Utils (#296)
- Eval Tests (#297)
- Support session token (#300)
- Dataset upload Enhancements (#292)
- Concept ID check befor model training (#295)
- RAG setup debug (#298)
- Requirements Update (#299)
- Model Upload v2 CLI (#269)
- Support Existing App in RAG (#275)
- Support RAG Prompter kwargs (#280)
- Custom Workflow id support in RAG (#291)
- Model Template Change in Model Train Test (#273)
- Dataset Upload summary fix (#282)
- Update Model Serving Docs (#287)
- Modified process_response_keys functions to fetch metadata info (#270)
- Assert user_id condition for RAG (#268)
- Changed demo link in README (#260)
- Fixed Mulitmodal input bug (#261)
- Workflow predict retry time to 10 minutes (#266)
- Update clarifai-grpc to 10.0.1 (#267)
- Test Cases for Model Upload (#256)
- Download Inputs functionality (#263)
- Added RAG base class (#262)
- RAG Chat Method (#264)
- RAG Upload Method (#265)
- Model upload examples moved to examples repo (#258)
- Use specific URL method for apps (#257)
- Loosen requirement constraints (#243)
- Update clarifai-grpc to 9.11.5
- Support Rank for PostInputsSearch (#255)
- CocoDetectionDataloader bug (#241)
- Codeql Change (#241)
- Seperate tests requiring secrets (#233)
- SDK Pending tasks (#232)
- add retry for workflow predict
- add constants for max inputs count in predict
- change annotation proto to bbox
- add search to README.md
- Add CHANGELOG.md
- Updated runner logic with parallel and error catching (#238)
- Removing internal_only Training Params (#231)
- Remove pytest requirement (#225)
- Remove omegaconf requirement (#235)
- Update clarifai-grpc to 9.11.0
- Support multimodal inputs for inference (#239)
- Ensure support for Python 3.10-3.12 (#226)
- Add MANIFEST.in back to include .css files
- Support Dataset Upload Status
- Support PAT as arg
- SDK cleanup(docs, examples, symlink to clarifai_utils, clarifai.auth)
- Refactor dataset upload process(loaders, dataloader)
- Fix Search top_k bug
- Model Training in SDK.
- Tests for Model Training.
- Fix base_url bug in passing while chained
- Moving Pycocotools requirement to extras(clarifai[all]).
- Support for model inference params
- PostInputsSearch Support
- Bump clarifai_grpc==9.10.0
- Pagination in listing
- Support list_annotations
- Supports custom metadata in dataloader, upload_from_csv
- Set clarifai_grpc to 9.8.1
- Reuse requirements.txt in setup.py
- Support Annotation Download
- Fix critical Version file not found bug in 9.9.1
- Reuse Version number from Version file
- Support Vector Search
- Workflow Create Bugs
- Support Workflow Create, Export
- Bump clarifai_grpc to 9.8.1
- Bump clarifai_grpc to 9.8.0
- Bump clarifai_grpc to 9.7.4
- Model Serving Support
- PyPi build issues