All pull requests

Pull requests

• [feat] Export a Router route as a standalone SentenceTransformer

  #4282·j341nono opened Oct 26, 2026Contributor·

  ·

• [feat] Add an evaluator for exporting sentence embeddings

  #4281·j341nono opened Oct 26, 2026Contributor·

  ·

• Docs: deployment guidance for cold starts and offline model loading

  #4276·anilkumara9 opened Oct 26, 2026·

  ·
  1

• [fix] Restore loss-owned weights when resuming from a checkpoint

  #4272·ViderStar opened Oct 25, 2026·

  ·

• Handle empty retrieval results in cross-encoder reranking evaluation

  #4264·Aurantium-47 opened Oct 23, 2026Contributor·

  ·
  1

• Document adaptive-layer evaluation during training

  #4261·Bembaby opened Oct 23, 2026Contributor·

  ·

• Infer ContrastiveTensionLoss labels for flattened inputs

  #4260·Bembaby opened Oct 23, 2026Contributor·

  ·

• [fix] Upcast int8/uint8 embeddings in the dense similarity functions

  #4259·nancy6547 opened Oct 23, 2026Contributor·

  ·

• Explain batch sizes for cached contrastive losses

  #4254·Bembaby opened Oct 22, 2026Contributor·

  ·

• Gather anchors for GISTEmbedLoss's anchor-anchor term across devices

  #4223·G26karthik opened Oct 18, 2026·

  ·
  1

• Export whole model with transformers.exporters

  #4178·echarlaix opened Oct 14, 2026Contributor·

  ·
  10

• Detect IR evaluator cutoff ties without full-size temporaries

  #4148·novain-shibu opened Oct 10, 2026·

  ·
  2

• Expose num_truncated_tokens in preprocess output (#4060)

  #4137·fruitstash opened Oct 7, 2026·

  ·

• Fix TSDAE reconstruction with left padding and shared EOS/PAD

  #4136·Afloat16 opened Oct 7, 2026Contributor·

  ·

• Exclude unretrieved positives from observed RRF ranking metrics

  #4114·Afloat16 opened Oct 2, 2026Contributor·

  ·

• Fix encode_document() alias selection while preserving explicit empty prompts

  #4095·PiUnknown opened Sep 29, 2026·

  ·
  1

  4

• Add opt-in lazy preprocessing for cached multimodal training

  #4091·palanigsharma182-ctrl opened Sep 27, 2026·

  ·

• [feat] Add min-max score normalization to DistillKLDivLoss family

  #4083·yjoonjang opened Sep 26, 2026Contributor·

  ·
  1

• [fix] Return per-input token embeddings from encode() with Flash Attention input flattening

  #4080·SulimanAbdulrazzaq opened Sep 25, 2026·

  ·
  1

• Cap reranked scores by the positives the first stage missed

  #4054·vineethsaivs opened Sep 20, 2026Contributor·

  ·
  1

• Docs: add production deployment guide for async servers (FastAPI/uvic…

  #3856·Shubh3005 opened Jul 12, 2026·

  ·
  1

• [feat] Add token-level sparse autoencoder modules

  #3796·robro612 opened Jun 4, 2026·

  ·
  1

• [feat] Implement Quantization Aware Training (QAT) Loss draft

  #3655·tomaarsen opened Feb 4, 2026Member·

  ·
  11

• Adding support for LMK pooling

  #3642·meetdoshi90 opened Jan 31, 2026·

  ·
  11

• Ensure encode() handles Series and other sequence types

  #3372·jteijema opened May 22, 2025·

  ·
  4