v0.3.1

What's Changed

• Update version.txt after 0.3.0 release by @loadams in #520
• Update supported model list by @tohtana in #519
• Enable streaming option in the OpenAI API server by @adk9 in #480
• Fix missing pydantic updates in legacy mii code by @loadams in #524
• Update docker container version by @loadams in #533
• Update CODEOWNERS by @loadams in #535
• Update labels to acquire new runners by @loadams in #534
• Update path triggers that were incorrect before by @loadams in #537
• Update clang-format version to match DeepSpeed by @loadams in #538

New Contributors

• @adk9 made their first contribution in #480

Full Changelog: v0.3.0...v0.3.1

DeepSpeed-MII v0.3.0

What's Changed

• Update version.txt by @loadams in #511
• Fix scheduling for non-persistent pipeline by @tohtana in #515
• Add Kubernetes health check route to REST server by @richiejp in #445
• Pydantic v2 migration by @mrwyattii in #423

New Contributors

• @richiejp made their first contribution in #445

Full Changelog: v0.2.4...v0.3.0

v0.2.4

What's Changed

• Update version.txt after 0.2.3 release by @mrwyattii in #434
• Update GH workflow and workflow runner requirements. by @loadams in #456
• Update pyzmq in requirements.txt by @ccoulombe in #447
• Add accelerate to requirements to improve MII-legacy model load times by @mrwyattii in #415
• Update create a PR workflow to latest version withh node js 20 fixes by @loadams in #461
• fix max_ragged_sequence_count check in _schedule_prompts by @dc3671 in #470
• Update mistral tests to fully open source version. by @loadams in #476
• Fix deprecation warning on escaped characters by @loadams in #478
• Remove Conversation from MII as it was deprecated and removed from transformers. by @loadams in #492
• Always Flush UIDs after Exceptions by @weiqisun in #491
• Allow model to generate added tokens - fix generation issue in Llama3 models by @weiqisun in #473
• Add explanations of MII code into comments by @mrwyattii in #493
• Update workflow task to use Ubuntu 22.04 by @loadams in #508
• Pin to use a specific version of transformers by @loadams in #510
• Update MII to switch from modelid to id by @loadams in #507
• Update model support by @mrwyattii in #429

New Contributors

• @ccoulombe made their first contribution in #447
• @dc3671 made their first contribution in #470
• @weiqisun made their first contribution in #491

Full Changelog: v0.2.3...v0.2.4

DeepSpeed-MII v0.2.3

What's Changed

• Update version.txt after 0.2.2 release by @mrwyattii in #413
• Add RTD by @mrwyattii in #321
• Add test for loading from local dir by @mrwyattii in #414
• Remove references to --extra-index-url in MII repo by @loadams in #421
• Add quantization config option by @mrwyattii in #433

Full Changelog: v0.2.2...v0.2.3

DeepSpeed-MII v0.2.2

What's Changed

• Update version.txt after 0.2.1 release by @mrwyattii in #404
• Fix for missing EOS token by @mrwyattii in #408
• text2img task to support negative prompts by @gauravrajguru in #407
• Disable model check in UT by @mrwyattii in #411
• fix: Fixed the issue where the mii.pipeline.pipe(stop) was ineffective by @kitstar in #409
• Add support for inpainting task in DS-MII by @gauravrajguru in #410

New Contributors

• @kitstar made their first contribution in #409

Full Changelog: v0.2.1...v0.2.2

DeepSpeed-MII v0.2.1

What's Changed

• Update version.txt after 0.2.0 release by @mrwyattii in #382
• Error on unknown generate fields by @mrwyattii in #386
• Make the order of outputs the same as the order of inputs when using mii.pipeline by @s-jse in #384
• Update landing page by @mrwyattii in #391
• Update CI workflows by @loadams in #392
• Fix recovery from deadlock by @tohtana in #394
• Readable token streaming support by @greshilov in #397
• Fix generate output order by @mrwyattii in #401
• Improve recovery from KV cache starvation by @tohtana in #403
• Adding OpenAI Compatible RESTful API by @PawanOsman in #317

New Contributors

• @s-jse made their first contribution in #384
• @greshilov made their first contribution in #397
• @PawanOsman made their first contribution in #317

Full Changelog: v0.2.0...v0.2.1

DeepSpeed-MII v0.2.0

New Features

• DeepSpeed-FastGen: Introducting Mixtral, Phi-2, and Falcon support with major performance and feature enhancements.

What's Changed

• Update version.txt after 0.1.3 release by @mrwyattii in #355
• Add pipeline unit tests by @mrwyattii in #359
• Update supported models list by @mrwyattii in #360
• Add restful_api_host into server args. by @sarattha in #364
• fix bug when mii_config is None by @mrwyattii in #367
• fix address already in use error on UT by @mrwyattii in #369
• Restrict when legacy unit tests are run by @mrwyattii in #370
• Make generate params pydantic model by @mrwyattii in #372
• Add model support unit test by @mrwyattii in #373
• Improve efficiency of scheduling and token sampiling by @tohtana in #377
• Update required DS version by @mrwyattii in #380

New Contributors

• @sarattha made their first contribution in #364

Full Changelog: v0.1.3...v0.2.0

DeepSpeed-MII v0.1.3

What's Changed

• Add RESTful API option for host by @mrwyattii in #348
• Fix for error messages in persistent deployment by @mrwyattii in #350
• Loosen unit test performance assert by @mrwyattii in #352
• Add device map config by @mrwyattii in #331

Full Changelog: v0.1.2...v0.1.3

DeepSpeed-MII v0.1.2

What's Changed

• Update RESTful API by @mrwyattii in #294
• Compatibility with DS Inference KV-cache flexibility PR by @cmikeh2 in #284
• Use smaller model for unit tests by @mrwyattii in #315
• Unify input/output types by @mrwyattii in #295
• Multi-process RESTful API by @mrwyattii in #328
• Add support for zero shot image classification task by @PhaniShekhar in #326
• Add safetensors support by @jihnenglin in #296
• Handling of edge case for text to image model by @gauravrajguru in #334
• fix PR creation in release workflow by @mrwyattii in #337

New Contributors

• @cmikeh2 made their first contribution in #284
• @PhaniShekhar made their first contribution in #326
• @jihnenglin made their first contribution in #296

Full Changelog: v0.1.1...v0.1.2

DeepSpeed-MII v0.1.1

What's Changed

• Add MII v0.1 unit tests by @mrwyattii in #262
• diffuser model load using model and path params by @gauravrajguru in #264
• Reorganize code structure, fix client import bug by @mrwyattii in #269
• Mitigate the risk of deadlock by @tohtana in #274
• Expose top-p, top-k, and temperature to generate APIs by @mrwyattii in #268
• Fix typo in README.md by @eltociear in #259
• Enable multi-prompt input for persistent deployments by @mrwyattii in #275
• Recompute when the deadlock is detected by @tohtana in #278
• prevent load_with_sys_mem when using stable diffusion by @mrwyattii in #279
• Add more generate() kwargs by @mrwyattii in #280

New Contributors

• @gauravrajguru made their first contribution in #264
• @eltociear made their first contribution in #259

Full Changelog: v0.1.0...v0.1.1