v0.14.3 Patch release

What's Changed

• Update version.txt after 0.14.2 release by @mrwyattii in #5458
• Add getter and setter methods for compile_backend across accelerators. by @vshekhawat-hlab in #5299
• Fix torch.compile error for PyTorch v2.3 by @tohtana in #5463
• Revert "stage3: efficient compute of scaled_global_grad_norm (#5256)" by @lekurile in #5461
• Update ds-chat CI workflow paths to include zero stage 1-3 files by @lekurile in #5462
• Update with ops not supported on Windows by @loadams in #5468
• fix: swapping order of parameters in create_dir_symlink method. by @alvieirajr in #5465
• Un-pin torch version in nv-torch-latest back to latest and skip test_compile_zero tests on v100 by @loadams in #5459
• re-introduce: stage3: efficient compute of scaled_global_grad_norm by @nelyahu in #5493
• Fix crash when creating Torch tensor on NPU with device=get_accelerator().current_device() by @harygo2 in #5464
• Fix compile wrapper by @BacharL in #5455
• enable phi3_mini autotp by @Yejing-Lai in #5501
• Fused adam for HPU by @BacharL in #5500
• [manifest] update mainfest to add hpp file in csrc. by @ys950902 in #5522
• enable phi2 autotp by @Yejing-Lai in #5436
• Switch pynvml to nvidia-ml-py by @loadams in #5529
• Switch from double quotes to match single quotes by @loadams in #5530
• [manifest] update mainfest to add hpp file in deepspeed. by @ys950902 in #5533
• New integration - CometMonitor by @alexkuzmik in #5466
• Improve _configure_optimizer() final optimizer log by @nelyahu in #5528
• Enhance testing: Skip fused_optimizer tests if not supported. by @vshekhawat-hlab in #5159
• Skip the UT cases that use unimplemented op builders. by @foin6 in #5372
• rocblas -> hipblas changes for ROCm by @rraminen in #5401
• Rocm warp size fix by @rraminen in #5402
• CPUAdam fp16 and bf16 support by @BacharL in #5409
• Optimize zero3 fetch params using all_reduce by @deepcharm in #5420
• Fix the TypeError for XPU Accelerator by @shiyang-weng in #5531
• Fix RuntimeError for moe on XPU: tensors found at least two devices by @shiyang-weng in #5519
• Remove synchronize calls from allgather params by @BacharL in #5516
• Avoid overwrite of compiled module wrapper attributes by @deepcharm in #5549
• Small typos in functions set_none_gradients_to_zero by @TravelLeraLone in #5557
• Adapt doc for #4405 by @oraluben in #5552
• Update to HF_HOME from TRANSFORMERS_CACHE by @loadams in #4816
• [INF] DSAttention allow input_mask to have false as value by @oelayan7 in #5546
• Add throughput timer configuration by @deepcharm in #5363
• Add Ulysses DistributedAttention compatibility by @Kwen-Chen in #5525
• Add hybrid_engine.py as path to trigger the DS-Chat GH workflow by @lekurile in #5562
• Update HPU docker version by @loadams in #5566
• Rename files in fp_quantize op from quantize.* to fp_quantize.* by @loadams in #5577
• [MiCS] Remove the handle print on DeepSpeed side by @ys950902 in #5574
• Update to fix sidebar over text by @loadams in #5567
• DeepSpeedCheckpoint: support custom final ln idx by @nelyahu in #5506
• Update minor CUDA version compatibility by @adk9 in #5591
• Add slide deck for meetup in Japan by @tohtana in #5598
• Fixed the Windows build. by @costin-eseanu in #5596
• estimate_zero2_model_states_mem_needs: fixing memory estiamtion by @nelyahu in #5099
• Fix cuda hardcode for inference woq by @Liangliang-Ma in #5565
• fix sequence parallel(Ulysses) grad scale for zero0 by @inkcherry in #5555
• Add Compressedbackend for Onebit optimizers by @Liangliang-Ma in #5473
• Updated hpu-gaudi2 tests content. by @vshekhawat-hlab in #5622
• Pin transformers version for MII tests by @loadams in #5629
• WA for Torch-compile-Z3-act-apt accuracy issue from the Pytorch repo by @NirSonnenschein in #5590
• stage_1_and_2: optimize clip calculation to use clamp by @nelyahu in #5632
• Fix overlap communication of ZeRO stage 1 and 2 by @penn513 in #5606
• fixes in _partition_param_sec function by @mmhab in #5613
• assumption of torch.initial_seed function accepting seed arg in DeepSpeedAccelerator abstract class is incorrect by @polisettyvarma in #5569
• pipe/_exec_backward_pass: fix immediate grad update by @nelyahu in #5605
• Monitor was always enabled causing performance degradation by @deepcharm in #5633

New Contributors

• @alvieirajr made their first contribution in #5465
• @harygo2 made their first contribution in #5464
• @alexkuzmik made their first contribution in #5466
• @foin6 made their first contribution in #5372
• @shiyang-weng made their first contribution in #5531
• @TravelLeraLone made their first contribution in #5557
• @oraluben made their first contribution in #5552
• @Kwen-Chen made their first contribution in #5525
• @adk9 made their first contribution in #5591
• @costin-eseanu made their first contribution in #5596
• @NirSonnenschein made their first contribution in #5590
• @penn513 made their first contribution in #5606

Full Changelog: v0.14.2...v0.14.3

v0.14.2 Patch release

What's Changed

• Update version.txt after 0.14.1 release by @mrwyattii in #5413
• Remove dtype(fp16) condition check for residual_add unit test by @raza-sikander in #5329
• [XPU] Use non_daemonic_proc by default on XPU device by @ys950902 in #5412
• Fix a convergence issues in TP topology caused by incorrect grad_norm. by @inkcherry in #5411
• Update 'create-pr' action in release workflow to latest by @loadams in #5415
• Update engine.py to avoid torch warning by @etiennebonnafoux in #5408
• Update _sidebar.scss by @fasterinnerlooper in #5293
• Add more tests into XPU CI by @Liangliang-Ma in #5427
• [CPU] Support SHM based inference_all_reduce in TorchBackend by @delock in #5391
• Add required paths to trigger AMD tests on PRs by @loadams in #5406
• Bug fix in split_index method by @bm-synth in #5292
• Parallel map step for DistributedDataAnalyzer map-reduce by @bm-synth in #5291
• Selective dequantization by @RezaYazdaniAminabadi in #5375
• Fix sorting of shard optimizer states files for universal checkpoint by @tohtana in #5395
• add device config env for the accelerator by @shiyuan680 in #5396
• 64bit indexing fused adam by @garrett4wade in #5187
• Improve parallel process of universal checkpoint conversion by @tohtana in #5343
• set the default to use set_to_none for clearing gradients in BF16 optimizer. by @inkcherry in #5434
• OptimizedLinear implementation by @jeffra in #5355
• Update README.md by @Jhonso7393 in #5453
• Update PyTest torch version to match PyTorch latest official (2.3.0) by @loadams in #5454

New Contributors

• @etiennebonnafoux made their first contribution in #5408
• @fasterinnerlooper made their first contribution in #5293
• @shiyuan680 made their first contribution in #5396
• @garrett4wade made their first contribution in #5187
• @Jhonso7393 made their first contribution in #5453

Full Changelog: v0.14.1...v0.14.2

v0.14.1 Patch release

What's Changed

• Update version.txt after 0.14.0 release by @mrwyattii in #5238
• Fp6 blog chinese by @xiaoxiawu-microsoft in #5239
• Add contributed HW support into README by @delock in #5240
• Set tp world size to 1 in ckpt load, if MPU is not provided by @samadejacobs in #5243
• Make op builder detection adapt to accelerator change by @delock in #5206
• Replace HIP_PLATFORM_HCC with HIP_PLATFORM_AMD by @rraminen in #5264
• Add CI for Habana Labs HPU/Gaudi2 by @loadams in #5244
• Fix attention mask handling in the Hybrid Engine Bloom flow by @deepcharm in #5101
• Skip 1Bit Compression and sparsegrad tests for HPU. by @vshekhawat-hlab in #5270
• Enabled LMCorrectness inference tests on HPU. by @vshekhawat-hlab in #5271
• Added HPU backend support for torch.compile tests. by @vshekhawat-hlab in #5269
• Average only valid part of the ipg buffer. by @BacharL in #5268
• Add HPU accelerator support in unit tests. by @vshekhawat-hlab in #5162
• Fix loading a universal checkpoint by @tohtana in #5263
• Add Habana Gaudi2 CI badge to the README by @loadams in #5286
• Add intel gaudi to contributed HW in README by @BacharL in #5300
• Fixed Accelerate Link by @wkaisertexas in #5314
• Enable mixtral 8x7b autotp by @Yejing-Lai in #5257
• support bf16_optimizer moe expert parallel training and moe EP grad_scale/grad_norm fix by @inkcherry in #5259
• fix comms dtype by @mayank31398 in #5297
• Modified regular expression by @igeni in #5306
• Docs typos fix and grammar suggestions by @Gr0g0 in #5322
• Added Gaudi2 CI tests. by @vshekhawat-hlab in #5275
• Improve universal checkpoint by @tohtana in #5289
• Increase coverage for HPU by @loadams in #5324
• Add NFS path check for default deepspeed triton cache directory by @HeyangQin in #5323
• Correct typo in checking on bf16 unit test support by @loadams in #5317
• Make NFS warning print only once by @HeyangQin in #5345
• resolve KeyError: 'PDSH_SSH_ARGS_APPEND' by @Lzhang-hub in #5318
• BF16 optimizer: Clear lp grads after updating hp grads in hook by @YangQun1 in #5328
• Fix sort of zero checkpoint files by @tohtana in #5342
• Add distributed_port for deepspeed.initialize by @LZHgrla in #5260
• [fix] fix typo s/simultanenously /simultaneously by @digger-yu in #5359
• Update container version for Gaudi2 CI by @raza-sikander in #5360
• compute global norm on device by @BacharL in #5125
• logger update with torch master changes by @rogerxfeng8 in #5346
• Ensure capacity does not exceed number of tokens by @jeffra in #5353
• Update workflows that use cu116 to cu117 by @loadams in #5361
• FP [6,8,12] quantizer op by @jeffra in #5336
• CPU SHM based inference_all_reduce improve by @delock in #5320
• Auto convert moe param groups by @jeffra in #5354
• Support MoE for pipeline models by @mosheisland in #5338
• Update pytest and transformers with fixes for pytest>= 8.0.0 by @loadams in #5164
• Increase CI coverage for Gaudi2 accelerator. by @vshekhawat-hlab in #5358
• Add CI for Intel XPU/Max1100 by @Liangliang-Ma in #5376
• Update path name on xpu-max1100.yml, add badge in README by @loadams in #5386
• Update checkout action on workflows on ubuntu 20.04 by @loadams in #5387
• Cleanup required_torch_version code and references. by @loadams in #5370
• Update README.md for intel XPU support by @Liangliang-Ma in #5389
• Optimize the fp-dequantizer to get high memory-BW utilization by @RezaYazdaniAminabadi in #5373
• Removal of cuda hardcoded string with get_device function by @raza-sikander in #5351
• Add custom reshaping for universal checkpoint by @tohtana in #5390
• fix pagable h2d memcpy by @GuanhuaWang in #5301
• stage3: efficient compute of scaled_global_grad_norm by @nelyahu in #5256
• Fix the FP6 kernels compilation problem on non-Ampere GPUs. by @JamesTheZ in #5333

New Contributors

• @vshekhawat-hlab made their first contribution in #5270
• @wkaisertexas made their first contribution in #5314
• @igeni made their first contribution in #5306
• @Gr0g0 made their first contribution in #5322
• @Lzhang-hub made their first contribution in #5318
• @YangQun1 made their first contribution in #5328
• @raza-sikander made their first contribution in #5360
• @rogerxfeng8 made their first contribution in #5346
• @JamesTheZ made their first contribution in #5333

Full Changelog: v0.14.0...v0.14.1

DeepSpeed v0.14.0

New Features

• DeepSpeed-FP6: The Power of FP6-Centric Serving for Large Language Models.

What's Changed

• Update version.txt after 0.13.5 release by @mrwyattii in #5229
• MOE gate fixes and enhancements by @mosheisland in #5156
• FP6 quantization end-to-end. by @loadams in #5234
• FP6 blog by @loadams in #5235

Full Changelog: v0.13.5...v0.14.0

v0.13.5 Patch release

What's Changed

• Update version.txt after 0.13.4 release by @mrwyattii in #5196
• Fix assertion to run pipeline engine with a compiled module by @tohtana in #5197
• Allow specifying MII branch on MII CI by @mrwyattii in #5208
• [zero++] Synchronize at the end of secondary partitioning and simplify the logic by @ByronHsu in #5216
• Add fp16 support of Qwen1.5 models (0.5B to 72B) to DeepSpeed-FastGen by @ZonePG in #5219
• Rename nv-torch-latest-cpu workflow to cpu-torch-latest by @loadams in #5226
• Fix moe cpu offload by @RezaYazdaniAminabadi in #5220
• Use deepspeed.comm instead of torch.distributed by @jinyouzhi in #5225
• fix fused_qkv model accuracy issue by @Yejing-Lai in #5217

Full Changelog: v0.13.4...v0.13.5

v0.13.4 Patch release

What's Changed

• Update version.txt after v0.13.3 release by @loadams in #5185
• Fixes for --extra-index-url by @loadams in #5183
• allow debug/experimental compiler backends by @tohtana in #5191
• Disable ninja by default by @mrwyattii in #5194
• [CPUAdam] Update full_precision_optimizer_states in docstring by @rohan-varma in #5181
• Add script to check for --extra-index-url by @loadams in #5184

New Contributors

• @rohan-varma made their first contribution in #5181

Full Changelog: v0.13.3...v0.13.4

v0.13.3 Patch release

What's Changed

• Update version.txt after 0.13.2 release by @mrwyattii in #5119
• Stop tracking backward chain of broadcast (ZeRO3) by @tohtana in #5113
• [NPU]ZeRO-Infinity feature compatibility by @misstek in #5077
• BF16 optimizer: Improve device utilization by immediate grad update by @deepcharm in #4975
• removed if condition in if collate_fn is None by @bm-synth in #5107
• disable compile tests for torch<2.1 by @mrwyattii in #5121
• Update inference test model names by @mrwyattii in #5127
• Fix issue with zero-sized file after merging file on curriculum map_reduce by @bm-synth in #5106
• Update return codes in PyTest to properly error out if tests fail by @loadams in #5122
• add missing methods to MPS_Accelerator by @mrwyattii in #5134
• Solve tensor vs numpy dtype conflicts in data efficiency map-reduce. by @bm-synth in #5108
• Fix broadcast deadlock for incomplete batches in data sample for data analysis by @bm-synth in #5117
• Avoid zero-sized microbatches for incomplete minibatches when doing curriculum learning by @bm-synth in #5118
• remove mandatory index key from output of metric_function in DataAnalysis map operation by @bm-synth in #5112
• tensorboard logging: avoid item() outside gas to improve performance by @nelyahu in #5135
• Check overflow on device without host synchronization for each tensor by @BacharL in #5115
• Update nv-inference torch version by @loadams in #5128
• Method run_map_reduce to fix errors when running run_map followed by run_reduce by @bm-synth in #5131
• Added missing isinstance check in PR 5112 by @bm-synth in #5142
• Fix UserWarning: The torch.cuda.*DtypeTensor constructors are no long… by @ShukantPal in #5018
• TestEmptyParameterGroup: replace fusedAdam with torch.optim.AdamW by @nelyahu in #5139
• Update deprecated HuggingFace function by @mrwyattii in #5144
• Pin to PyTest 8.0.0 by @loadams in #5163
• get_grad_norm_direct: fix a case of empty norm group by @nelyahu in #5148
• Distributed in-memory map-reduce for data analyzer by @bm-synth in #5129
• DeepSpeedZeroOptimizer_Stage3: remove cuda specific optimizer by @nelyahu in #5138
• MOE: Fix save checkpoint when TP > 1 by @mosheisland in #5157
• Fix gradient clipping by @tohtana in #5150
• Use ninja to speed up build by @jinzhen-lin in #5088
• Update flops profiler to handle attn and matmul by @KimmiShi in #4724
• Fix allreduce for BF16 and ZeRO0 by @tohtana in #5170
• Write multiple items to output file at once, in distributed data analyzer. by @bm-synth in #5169
• Fix typos in blogs/ by @jinyouzhi in #5172
• Inference V2 Human Eval by @lekurile in #4804
• Reduce ds_id name length by @jomayeri in #5176
• Switch cpu-inference workflow from --extra-index-url to --index-url by @loadams in #5182

New Contributors

• @bm-synth made their first contribution in #5107
• @KimmiShi made their first contribution in #4724

Full Changelog: v0.13.2...v0.13.3

v0.13.2 Patch release

What's Changed

• Update version.txt after 0.13.1 release by @mrwyattii in #5002
• Support exclude_frozen_parameters for save_16bit_model by @LZHgrla in #4999
• Allow nightly tests dispatch by @mrwyattii in #5014
• Enable hpz based on secondary tensor presence by @HeyangQin in #4906
• Enable workflow dispatch on all workflows by @loadams in #5016
• [minor] improve code quality and readablilty by @ByronHsu in #5011
• Update falcon fused type order by @Yejing-Lai in #5007
• Fix error report of DSElasticAgent._set_master_addr_port() by @RobinDong in #4985
• DS #4993 #662 : autotune single node hostfile bugfix by @oushu1zhangxiangxuan1 in #4996
• [minor] Improve logging for multiprocesses by @ByronHsu in #5004
• deepspeed/launcher: add launcher_helper as each rank's start portal by @YizhouZ in #4699
• Graph capture support on HPU accelerators by @deepcharm in #5013
• launcher/launcher_helper.py: fix PMI name and add EnvironmentError by @YizhouZ in #5025
• Remove MI100 badge from landing page by @mrwyattii in #5036
• Remove coverage reports from workflows and fix for inference CI by @loadams in #5028
• Remove Megatron-DeepSpeed CI workflow by @mrwyattii in #5038
• Fix P40 CI failures by @mrwyattii in #5037
• Fix for nightly torch CI by @mrwyattii in #5039
• Fix nv-accelerate and nv-torch-latest-v100. by @loadams in #5035
• update inference pages to point to FastGen by @mrwyattii in #5029
• launcher_helper: enable fds passing by @YizhouZ in #5042
• Fix nv-torch-latest-cpu CI by @mrwyattii in #5045
• [NPU] Add NPU to support hybrid engine by @CurryRice233 in #4831
• MoE type hints by @ringohoffman in #5043
• [doc] update inference related docs from mp_size to tensor_parallel for TP by @yundai424 in #5048
• Fix broken model names in inference CI by @mrwyattii in #5053
• [NPU] Change log level to debug by @CurryRice233 in #5051
• Delay reduce-scatter for ZeRO3 leaf modules by @tohtana in #5008
• Optimize grad_norm calculations by reducing device/host dependency by @nelyahu in #4974
• load linear layer weight with given dtype by @polisettyvarma in #4044
• Update import for changes to latest diffusers by @mrwyattii in #5065
• adding hccl to init_distributed function description by @nelyahu in #5034
• [Zero++ qgZ] Fall back to reduce_scatter if tensor.numel() % (2 * global_world_size) != 0 by @ByronHsu in #5056
• Make batch size documentation clearer by @segyges in #5072
• [doc/1-line change] default stage3_param_persistence_threshold is wrong in the doc by @ByronHsu in #5073
• Further refactor deepspeed.moe.utils + deepspeed.moe.layer type hints by @ringohoffman in #5060
• Fix verification for ZeRO3 leaf module by @tohtana in #5074
• Stop tracking backward chain of broadcast in initialization by @tohtana in #5075
• Update torch version for nv-torch-latest-cpu by @loadams in #5086
• Add backwards compatibility w/ older versions of diffusers (<0.25.0) by @lekurile in #5083
• Enable torch.compile with ZeRO (Experimental) by @tohtana in #4878
• Update nv-accelerate to latest torch by @loadams in #5040
• HPU Accelerator: fix supported_dtypes API by @nelyahu in #5094
• [NPU] replace 'cuda' with get_accelerator().device_name() by @minchao-sun in #5095
• optimize clip_grad_norm_ function by @mmhab in #4915
• [xs] fix ZEROPP convergence test by @yundai424 in #5061
• Switch hasattr check from compile to compiler by @loadams in #5096
• Split is_synchronized_device api to multiple apis by @BacharL in #5026
• 47% FastGen speedup for low workload - refactor allocator by @HeyangQin in #5090
• Support exclude_frozen_parameters for zero_to_fp32.py script by @andstor in #4979
• Fix alignment of optimizer states when loading by @tohtana in #5105
• Skip Triton import for AMD by @lekurile in #5110
• Add HIP conversion file outputs to .gitignore by @lekurile in #5111
• Remove optimizer step on initialization by @tohtana in #5104

New Contributors

• @ByronHsu made their first contribution in #5011
• @RobinDong made their first contribution in #4985
• @oushu1zhangxiangxuan1 made their first contribution in #4996
• @yundai424 made their first contribution in #5048
• @segyges made their first contribution in #5072
• @andstor made their first contribution in #4979

Full Changelog: v0.13.1...v0.13.2

v0.13.1 Patch release

What's Changed

• Update version.txt after 0.13.0 release by @mrwyattii in #4982
• Update FastGen blog title by @arashb in #4983
• Fix the MoE-params gradient-scaling by @RezaYazdaniAminabadi in #4957
• fix some typo under blogs/ by @digger-yu in #4988
• Fix placeholder value in FastGen Blog by @mrwyattii in #5000
• fix for DS_ENV issue by @jeffra in #4992
• Delete unused --deepspeed_mpi command line argument by @ShukantPal in #4981
• Make installable without torch by @mrwyattii in #5001
• Implement some APIs of HPU accelerator by @mmhab in #4935
• Refactor the Qwen positional emebdding config code by @ZonePG in #4955

Full Changelog: v0.13.0...v0.13.1

DeepSpeed v0.13.0

New Features

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

What's Changed

• Update version.txt after 0.12.6 release by @mrwyattii in #4850
• doc corrections by @goodship1 in #4861
• Fix exception handling in get_all_ranks_from_group() function by @HeyangQin in #4862
• deepspeed engine: fp16 support validation on init by @nelyahu in #4843
• Remove hooks on gradient accumulation on engine/optimizer destroy by @chiragjn in #4858
• optimize grad_norm calculation in stage3.py by @mmhab in #4436
• Fix f-string messages by @li-plus in #4865
• [NPU] Fix npu offload bug by @CurryRice233 in #4883
• Partition parameters: Minor refactoring of use_secondary_tensor condition by @deepcharm in #4868
• Pipeline: Add support to eval micro bs configuration by @nelyahu in #4859
• zero_to_fp32.py: Handle a case where shape doesn't have numel attr by @nelyahu in #4842
• Add support of Microsoft Phi-2 model to DeepSpeed-FastGen by @arashb in #4812
• Support cpu tensors without direct device invocation by @abhilash1910 in #3842
• add sharded loading for safetensors in AutoTP by @sywangyi in #4854
• [XPU] XPU accelerator support for Intel GPU device by @delock in #4547
• enable starcode((kv_head=1)) autotp by @Yejing-Lai in #4896
• Release overlap_comm & contiguous_gradients restrictions for ZeRO 1 by @li-plus in #4887
• [NPU]Add ZeRO-Infinity feature for NPU by @misstek in #4809
• fix num_kv_heads sharding in uneven autoTP for Falcon-40b by @Yejing-Lai in #4712
• Nvme offload checkpoint by @eisene in #4707
• Add WarmupCosineLR to Read the Docs by @dwyatte in #4916
• Add Habana Labs HPU accelerator support by @deepcharm in #4912
• Unit tests for MiCS by @zarzen in #4792
• Fix SD workflow to work with latest diffusers version by @lekurile in #4918
• [Fix] Fix cpu inference UT failure by @delock in #4430
• Add paths to run SD tests by @loadams in #4919
• Change PR/schedule triggers for CPU-inference by @loadams in #4924
• fix falcon-40b accuracy issue by @Yejing-Lai in #4895
• Refactor the positional emebdding config code by @arashb in #4920
• Pin to triton 2.1.0 to fix issues with nv-inference by @loadams in #4929
• Add support of Qwen models (7b, 14b, 72b) to DeepSpeed-FastGen by @ZonePG in #4913
• DeepSpeedZeroOptimizer: refactor bit16 flattening to support more accelerators by @nelyahu in #4833
• Fix confusing width in simd_load by @yzhblind in #4714
• Specify permissions for secrets.GITHUB_TOKEN by @mrwyattii in #4927
• Enable quantizer op on ROCm by @rraminen in #4114
• autoTP for Qwen by @inkcherry in #4902
• Allow specifying mii branch for nv-a6000 workflow by @mrwyattii in #4936
• Only run MII CI for inference changes by @mrwyattii in #4939
• InfV2 - remove generation config requirement by @mrwyattii in #4938
• Cache HF model list for inference tests by @mrwyattii in #4940
• Fix docs inconsistency on default value for ignore_unused_parameters by @loadams in #4949
• Fix bug in CI model caching by @mrwyattii in #4951
• fix uneven issue & add balance autotp by @Yejing-Lai in #4697
• Optimize preprocess for ragged batching by @tohtana in #4942
• Fix bug where ZeRO2 never uses the reduce method. by @CurryRice233 in #4946
• [docs] Add new autotp supported model in tutorial by @delock in #4960
• Add missing op_builder.hpu component for HPU accelerator by @nelyahu in #4963
• Stage_1_and_2.py: fix assert for reduce_scatter configurations combinations by @nelyahu in #4964
• [MiCS]Add the path to support sequence_data_parallel on MiCS by @ys950902 in #4926
• Update the DeepSpeed Phi-2 impl. to work with the HF latest changes by @arashb in #4950
• Prevent infinite recursion when DS_ACCELERATOR is set to cuda by @ShukantPal in #4962
• Fixes for training models with bf16 + freshly initialized optimizer via load_module_only by @haileyschoelkopf in #4141
• params partition for skip_init by @inkcherry in #4722
• Enhance query APIs for text generation by @tohtana in #4965
• Add API to set a module as a leaf node when recursively setting Z3 hooks by @tohtana in #4966
• Fix T5 and mistral model meta data error by @Yejing-Lai in #4958
• FastGen Jan 2024 blog by @mrwyattii in #4980

New Contributors

• @chiragjn made their first contribution in #4858
• @li-plus made their first contribution in #4865
• @misstek made their first contribution in #4809
• @dwyatte made their first contribution in #4916
• @ZonePG made their first contribution in #4913
• @yzhblind made their first contribution in #4714
• @ShukantPal made their first contribution in #4962
• @haileyschoelkopf made their first contribution in #4141

Full Changelog: v0.12.6...v0.13.0