1 Release published by 1 person

• v3.1.1 v3.1.1 - Patch hard negative mining & remove `numpy<2` restriction

  published Sep 19, 2024

3 Pull requests merged by 1 person

• [fix] Ensure that the embeddings from hard negative mining are normalized

  #2944 merged Sep 19, 2024

• [metadata] Extend pyproject.toml metadata

  #2943 merged Sep 19, 2024

• [deps] Attempt to remove numpy restrictions

  #2937 merged Sep 17, 2024

3 Pull requests opened by 3 people

• Add negatives in CachedGISTEmbedLoss

  #2946 opened Sep 20, 2024

• [ENH] -- `CrossEncoder.rank`

  #2947 opened Sep 20, 2024

• [feat] Add option to mask prompts with left-padded tokenizer and corpus and query prompts to IREvaluator

  #2951 opened Sep 23, 2024

4 Issues closed by 3 people

• Necessity of torchvision dependency

  #1644 closed Sep 23, 2024

• 'SentenceTransformer' object has no attribute 'get_input_embeddings'

  #2949 closed Sep 23, 2024

• cannot import name 'is_torch_npu_available' from 'transformers'

  #2895 closed Sep 19, 2024

• Version 3.1.0 forces a downgrade to Numpy 1.x

  #2933 closed Sep 17, 2024

8 Issues opened by 8 people

• MNRL with Multiple hard negatives per query and NoDuplicatesBatchSampler

  #2954 opened Sep 23, 2024

• CROSS-ENCODER RANK ERROR

  #2953 opened Sep 23, 2024

• Releases on conda-forge via the sentence-transformers-feedstock

  #2952 opened Sep 23, 2024

• AttributeError: 'LatentAttentionConfig' object has no attribute '_attn_implementation_internal'

  #2950 opened Sep 22, 2024

• Limiting CPU Usage and Optimizing Offline Encoding with Sentence Transformer

  #2948 opened Sep 21, 2024

• Best practice to train on multiple datasets with different prompts

  #2945 opened Sep 19, 2024

• raise AttributeError(f"'{type(self).__name__}' object has no attribute '{name}'") AttributeError: 'SentenceTransformer' object has no attribute 'model_body'

  #2942 opened Sep 18, 2024

• Learning from padded values during model training when generating token_embeddings

  #2941 opened Sep 18, 2024

5 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.

• Train and embed with pretokenized input?

  #2935 commented on Sep 18, 2024 • 0 new comments

• Increase the number of negatives in CachedGISTEmbedLoss to 2

  #2932 commented on Sep 19, 2024 • 0 new comments

• Please future prove `clean_up_tokenization_spaces`

  #2922 commented on Sep 20, 2024 • 0 new comments

• Update conda-forge feedstock to include missing dependencies

  #1279 commented on Sep 23, 2024 • 0 new comments

• # Pass Image color channels information to Transformers

  #2846 commented on Sep 20, 2024 • 0 new comments