26 Pull requests merged by 20 people

• Fix the issue with --load_quantized_model_with_autoawq

  #1747 merged Feb 7, 2025

• Implement baselines as a fixture and with simple rebase support

  #1732 merged Feb 7, 2025

• Enable the i2vgen pipeline

  #1670 merged Feb 7, 2025

• Add _prepare_inputs_for_generation

  #1743 merged Feb 7, 2025

• Add diffuser tests for optimized sdxl flow on HPU

  #1554 merged Feb 7, 2025

• Add requirements.txt to sentence transformer training sts

  #1754 merged Feb 7, 2025

• Adds requirements.txt to sentence transformers training paraphrases

  #1753 merged Feb 7, 2025

• fix llama FP8 perf issue, kvcache.update should be used since FP8 pat…

  #1756 merged Feb 7, 2025

• Change clip-roberta/bridgetower not to use fast_ddp

  #1749 merged Feb 7, 2025

• Adjust baselines for lower number of epochs improved perplexity, lowe…

  #1748 merged Feb 7, 2025

• Accuracy fix for llama3.1-70B in eager/torch.compile mode

  #1746 merged Feb 7, 2025

• Add batch splitting in attention layer to hide NIC latency(#14)

  #1640 merged Feb 7, 2025

• Clean up README examples

  #1709 merged Feb 6, 2025

• Add the inline_inbuilt_nn_modules option

  #1617 merged Feb 6, 2025

• Add batch dim idx to support latest deepspeed DistributedAttention

  #1725 merged Feb 6, 2025

• add video-llava model support

  #1522 merged Feb 5, 2025

• [SW_212175] FLAN-T5 has bad performance when using regional compilation

  #1744 merged Feb 5, 2025

• Supports Bitsandbytes development on HPU (#117)

  #1714 merged Feb 5, 2025

• Fix the incorrect output of sdxl inpaint

  #1737 merged Feb 5, 2025

• edit mixtral quantization config file (#114)

  #1739 merged Feb 5, 2025

• Resolve 'NoneType' object has no attribute 'gate_proj' err when applying EP in DeepSeek-V2

  #1740 merged Feb 5, 2025

• Fix graph breaks in Mixtral (#65)

  #1705 merged Feb 5, 2025

• Add dynamo cache size limit option

  #1619 merged Feb 5, 2025

• Update DS config to align with recommended settings

  #1730 merged Feb 5, 2025

• Removing HL_DS_DISTRIBUTED_ATTENTION_SEQ_DIM as it's not needed from 1.20

  #1726 merged Feb 5, 2025

• Update README.md - correction in diffusers example

  #1742 merged Feb 4, 2025

11 Pull requests opened by 9 people

• Upstream Accelerate

  #1741 opened Feb 4, 2025

• Enable Llama 3.1 405B in FP8 (#124)

  #1745 opened Feb 5, 2025

• fix dpo crash in transformers 4.48

  #1750 opened Feb 6, 2025

• fix the output length in image_to_text test to make sure the outpu…

  #1751 opened Feb 6, 2025

• Add safe_globals to resume training on PyTorch 2.6

  #1752 opened Feb 6, 2025

• Fix Experts Indexing in MoE for Mixtral: Align experts_max with Number of Available Experts

  #1755 opened Feb 6, 2025

• Enable attention selection for wav2vec2

  #1757 opened Feb 7, 2025

• fix dependency issue with --load_quantized_model_with_autoawq

  #1759 opened Feb 8, 2025

• Fix for Falcon image-to-text crash

  #1760 opened Feb 8, 2025

• [SW-208588] Add HPU fp8 Dynamic MOE (#93)

  #1761 opened Feb 8, 2025

• [SW-216693]: UX code change

  #1764 opened Feb 10, 2025

1 Issue opened by 1 person

• Flux model training not learning my image features

  #1763 opened Feb 9, 2025

16 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• support llava1.5 lora finetuning.

  #1487 commented on Feb 10, 2025 • 5 new comments

• Diffusers: Simplified the README files. Updated CI tests.

  #1718 commented on Feb 9, 2025 • 4 new comments

• Add GLM4V

  #1668 commented on Feb 10, 2025 • 2 new comments

• Extend lm_eval functionality

  #1729 commented on Feb 7, 2025 • 2 new comments

• Update Text Generation run instructions for 1.19 release

  #1734 commented on Feb 4, 2025 • 1 new comment

• LlavaNextForConditionalGeneration.forward() got an unexpected keyword argument 'token_idx'

  #1708 commented on Feb 5, 2025 • 0 new comments

• add boft support in stable-diffusion

  #1295 commented on Feb 3, 2025 • 0 new comments

• add cogvideox support for gaudi.

  #1600 commented on Feb 8, 2025 • 0 new comments

• [Ecosystem] enable saving and loading FP8 model(#53)

  #1683 commented on Feb 7, 2025 • 0 new comments

• Added Mamba model using kernel to improve the performance

  #1689 commented on Feb 6, 2025 • 0 new comments

• Upgrade to Transformers v4.48

  #1698 commented on Feb 7, 2025 • 0 new comments

• add requirements file for no-deps

  #1711 commented on Feb 6, 2025 • 0 new comments

• Enabling Snowflake Arctic on Gaudi 3

  #1719 commented on Feb 5, 2025 • 0 new comments

• Fix training crash issue on multi-nodes when dataloader_num_workers>0

  #1721 commented on Feb 4, 2025 • 0 new comments

• DeepSeek_v3 support

  #1735 commented on Feb 10, 2025 • 0 new comments

• Added sliding window feature to Giudi Gemma2 model

  #1736 commented on Feb 5, 2025 • 0 new comments