Search results

Substantial numerical differences with Minimax MSA attention vs Minimax's Reference

bug

Status: Open.

#46762 In huggingface/transformers;

· CoffeeVampir3 opened on Jun 19, 2026

FR: Implement Downstream Integration Smoke Tests in CI/CD to Prevent Ecosystem Regression

Feature requestRequest for a new featureRequest for a new feature

Status: Open.

#46756 In huggingface/transformers;

· briancullinan2 opened on Jun 18, 2026

apply_chat_template crashes with IndexError when passed an empty list

bug

Status: Open.

#46752 In huggingface/transformers;

· sharmax-vikas opened on Jun 18, 2026

LoRA fine-tuning of an FP8 checkpoint is blocked: get_peft_model does not clear validate_quantization_for_training (QuantizationMethod.FP8)

Status: Open.

#46736 In huggingface/transformers;

· brunopistone opened on Jun 17, 2026

LabelSmoother ignores num_items_in_batch, inflating gradients by ~gradient_accumulation_steps with label smoothing

Status: Open.

#46731 In huggingface/transformers;

· Incheonkirin opened on Jun 17, 2026

Aria: get_number_of_image_patches returns wrong count with max_image_size=980 when best resolution has odd 490 multiples

Status: Open.

#46728 In huggingface/transformers;

· arnavkewalram opened on Jun 17, 2026

Several models hardcode torch.float64 in their forward pass, crashing on MPS (Apple Silicon)

Status: Open.

#46723 In huggingface/transformers;

· qflen opened on Jun 17, 2026

Clarify Gemma4 vision bidirectional mask behavior for full vs sliding attention

Status: Open.

#46711 In huggingface/transformers;

· wnma3mz opened on Jun 17, 2026

Accuracy regression in DeepSeek-R1-Distill-Llama-8B after upgrading Transformers from 4.55 to 5.9

bug

Status: Open.

#46710 In huggingface/transformers;

· xiaowangintel opened on Jun 17, 2026

[DiffusionGemmaGenerationOutput] image-text-to-text pipeline does not work

bug

Status: Open.

#46701 In huggingface/transformers;

· KoichiYasuoka opened on Jun 17, 2026

Swin2SR image processor over-pads images when dimensions are already divisible by size_divisor

Status: Open.

#46697 In huggingface/transformers;

· arnavkewalram opened on Jun 16, 2026

Performance regression with FA2 and long-context generation after 5.2.0

bug

Status: Open.

#46693 In huggingface/transformers;

· andreasgoulas opened on Jun 16, 2026