sentence-transformers · Issues· 1303 open

Cached losses (GradCache) trigger one DDP all-reduce per mini-batch per column instead of one per step

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

Finetuning Cross-Encoder model results in constant prediction but when validated on test set it shows 100 F1 and accuracy by evaluator function

Explicit device assignment ignored in multi-GPU (PyTriton) environment

The special_tokens in tokenizer should also be controlled by do_lower_case in encoder_config.

Mixing modalities in `encode()` doesn't work for CLIP

model.safetensors is missing read permissions after saving with SentenceTransformer.save()

Save individual routes from a Router-based model as standalone artifacts

No constructor-based way to set query_length/document_length on `MultiVectorEncoder`

Any pretrained models available for commercial use?

Problems in training multi-modal models with large batch sizes

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

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

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

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