Lightning-habana

Added

- Added support for Intel Gaudi Profiler. Deprecate `HABANA_PROFILE` environment variable from HPUProfiler. ([158](https://github.com/Lightning-AI/lightning-Habana/pull/158))
- Added support for FP8 inference. ([162](https://github.com/Lightning-AI/lightning-Habana/pull/162))
- Added support for LightningCLI. ([173](https://github.com/Lightning-AI/lightning-Habana/pull/173))
- Added experimental support for FSDP on HPU. ([174](https://github.com/Lightning-AI/lightning-Habana/pull/174))
- Added support for FP8 inference with DeepSpeed. ([176](https://github.com/Lightning-AI/lightning-Habana/pull/176))
- Updated the lightning version check for using FSDP. ([182](https://github.com/Lightning-AI/lightning-Habana/pull/182))


Changed

- Changed HPUParallelStrategy to HPUDDPStrategy ([160](https://github.com/Lightning-AI/lightning-Habana/pull/160))
- Changed HPU docker image based on Synapse AI release 1.15.0 ([166](https://github.com/Lightning-AI/lightning-Habana/pull/166))
- Updated to Intel Gaudi software Release 1.15.1 ([171](https://github.com/Lightning-AI/lightning-Habana/pull/171))

Fixed

- Fixed "No profiler activity found" error with HPUProfiler. ([172](https://github.com/Lightning-AI/lightning-Habana/pull/172))

Removed

-

Deprecated

-

Added

- Added DeepSpeed precision plugin for HPU ([147](https://github.com/Lightning-AI/lightning-Habana/pull/147))
- Added support for fp8 training. ([149](https://github.com/Lightning-AI/lightning-Habana/pull/149))

Changed

- Decoupled return strings of firmware, synapse version helper ([137](https://github.com/Lightning-AI/lightning-Habana/pull/137))
- Changed HPU docker image based on Synapse AI release 1.14.0 ([140](https://github.com/Lightning-AI/lightning-Habana/pull/140))

Fixed

- Fixed fabric imports for HPU strategies ([126](https://github.com/Lightning-AI/lightning-Habana/pull/126))
- Enabling tests and examples of fabric with HPU ([139](https://github.com/Lightning-AI/lightning-Habana/pull/139))
- Fixes an API break due to non-strict loading in Trainer ([150](https://github.com/Lightning-AI/lightning-Habana/pull/150))

Removed

-

Deprecated

- `aot_hpu_training_backend` will be deprecated. Use `hpu_backend` instead for torch compile with hpu ([148](https://github.com/Lightning-AI/lightning-Habana/pull/148))

Added

- Added support for Deepspeed inference on HPU with tests and documentation ([110](https://github.com/Lightning-AI/lightning-Habana/pull/110))
- Added tests, examples, and documentation for dynamic shapes with recipe caching ([107](https://github.com/Lightning-AI/lightning-Habana/pull/107))
- Added preview of torch compile with tests and documentation ([119](https://github.com/Lightning-AI/lightning-Habana/pull/119))

Changed

- Changed HPU docker image based on Synapse AI release 1.13.0 ([114](https://github.com/Lightning-AI/lightning-Habana/pull/114))
-

Fixed

- Fixed fabric imports for HPU strategies ([126](https://github.com/Lightning-AI/lightning-Habana/pull/126))

Removed

-

Deprecated

-

Added

- Added tests, examples and documentation for HPUPrecisionPlugin with autocast ([94](https://github.com/Lightning-AI/lightning-Habana/pull/94))
- Added test to validate checkpoint resuming with HPUDeepSpeedStrategy ([95](https://github.com/Lightning-AI/lightning-Habana/pull/95))
- Added support for lightning 2.1 ([100](https://github.com/Lightning-AI/lightning-Habana/pull/100), [#105](https://github.com/Lightning-AI/lightning-Habana/pull/105))

Changed

- Changed HPU docker image based on synapse AI release 1.12.0 ([90](https://github.com/Lightning-AI/lightning-Habana/pull/90))
- Use standard API's and Remove env variable to get HPU distributed backend ([91](https://github.com/Lightning-AI/lightning-Habana/pull/91))
- Changed HPU docker image based on synapse AI release 1.12.1, updated hooks ([106](https://github.com/Lightning-AI/lightning-Habana/pull/106))


Fixed

-


Removed

-


Deprecated

-

Added

- Documentation with examples for using DeepSpeed with HPU ([64](https://github.com/Lightning-AI/lightning-Habana/pull/64))
- Add autocast using HPUPrecision plugin ([66](https://github.com/Lightning-AI/lightning-Habana/pull/66), [#75](https://github.com/Lightning-AI/lightning-Habana/pull/75))
- Demonstrate HPU Graphs support ([67](https://github.com/Lightning-AI/lightning-Habana/pull/67))
- Enhance test coverage of DeepSpeed strategy on HPU ([68](https://github.com/Lightning-AI/lightning-Habana/pull/68))
- Added version check helper to use right release ([75](https://github.com/Lightning-AI/lightning-Habana/pull/75), [#76](https://github.com/Lightning-AI/lightning-Habana/pull/76))
- Implement reduce with parallel plugin ([77](https://github.com/Lightning-AI/lightning-Habana/pull/77))

Changed

- Changed HPU docker image based on synapse AI release 1.11.0 & upgraded deepspeed plugin to version 0.9.4 ([61](https://github.com/Lightning-AI/lightning-Habana/pull/61))

Fixed

- Fixed optimizer priority based on deepspeed specification ([36](https://github.com/Lightning-AI/lightning-Habana/pull/69))
- Fixed missing extras in package ([70](https://github.com/Lightning-AI/lightning-Habana/pull/70))

Deprecated

- Warn on HMP deprecation from `HPUPrecision` plugin ([65](https://github.com/Lightning-AI/lightning-Habana/pull/65))

Added

- Added tests for mixed precision training ([36](https://github.com/Lightning-AI/lightning-Habana/pull/36))
- Example to include mixed precision training ([54](https://github.com/Lightning-AI/lightning-Habana/pull/54))

Changed

- Enabled skipped tests based on registered strategy, accelerator ([46](https://github.com/Lightning-AI/lightning-Habana/pull/46))

Fixed

- Fixed Attribute Error ([43](https://github.com/Lightning-AI/lightning-Habana/pull/43))
- Fixed wrong imports ([44](https://github.com/Lightning-AI/lightning-Habana/pull/44))
- Fixed graph breaks in test/val phases in lazy mode ([45](https://github.com/Lightning-AI/lightning-Habana/pull/45))