Repository Issue Activity (beta)

#3980 Loading a LoRA CrossEncoder checkpoint does not restore a trained `modules_to_save` classification head

#3973 Backend ignores root level ONNX files, and cannot detect full model exports

#3953 Let a module declare an extra output type (e.g. multi-vector + sparse from one model)

#3348 NoDuplicatesBatchSampler __len__ method returns higher number of batches than the actual one

#3950 [`docs`] 'Performance ratio' in efficiency.rst GPU/CPU performance images isn't clearly about NDCG@10

#3939 `MultiVectorInformationRetrievalEvaluator` keeps the first model’s similarity function when reused

#3247 Loading a PEFT model uses `peft.peft_model.PeftModelForFeatureExtraction`; incompatible with ST PEFT Methods

#3438 Not able to get optimized onnx model weights file

#3929 paraphrase_mining_embeddings keeps max_pairs - 1 candidates

#3923 Docs: deployment guidance for cold starts and offline model loading (scale-to-zero, local_files_only, warmup)

#3357 Add a utility to easily find the underlying `PreTrainedModel` subclass

#3894 Similarity and cross-encoder scores computed below float32 create spurious ties in low-precision inference

#3916 CrossEncoder PEFT fine-tuning: confusing TypeError (BatchEncoding not Tensor) with manual get_peft_model; no official add_adapter example

#3903 mine_hard_negatives crashes on a tensor shape when num_negatives exceeds the range_min/range_max window

#3908 `RuntimeError: _apply(): Couldn't swap Linear.weight` when `torchao` with `torch.compile`

#3899 community_detection truncates a community at sort_max_size when similarities equal the threshold

#3881 CrossEncoder silently drops the whole text pair when the backbone's chat template ignores the query/document roles

#3757 Question regarding KL Divergence gradient flow in `AdaptiveLayerLoss` (edit: and possible issue with layers weighting schema)

#3885 Feature request for encoder-decoder reranker

#3889 Add the KaLM-Reranker-V1 family to the Sentence Transformers v5.4 integrations collection

#3896 sentence-transformers>=5.0 hard-imports torchcodec at package load, crashing text-only usage when torchcodec/FFmpeg is broken

#3874 Per-sample `video_metadata` gets attached to the wrong video when a batch mixes dict-wrapped and raw videos

#3870 mine_hard_negatives: cache_folder key ignores query_prompt/corpus_prompt, silently returns stale embeddings

#3884 Architectural Review of test_transformer.py

#3206 OSError from bad request between sentence-transformers and huggingface_hub