Search results

conv1d with large groups (depthwise) is extremely slow on CUDA — O(groups) kernel launches

Status: Open.

#3389 In huggingface/candle;

· m96-chan opened on Feb 27, 2026

Metal SDPA kernel produces NaN with non-square mask + GQA

Status: Open.

#3388 In huggingface/candle;

· starpit opened on Feb 26, 2026

Subject: Contribution for Metal Device Initialization Fix in Candle

Status: Open.

#3369 In huggingface/candle;

· PlutoEternity opened on Feb 15, 2026

Error: Get error "forward compatibility was attempted on non supported HW" on RTX 5090.

Status: Open.

#3358 In huggingface/candle;

· tjliupeng opened on Feb 3, 2026

linear_split in candle-metal-kernels cannot handle empty tensors

Status: Open.

#3353 In huggingface/candle;

· SKRohit opened on Feb 1, 2026

PaddleOCR-VL consider support bf16 or f16 inference

Status: Open.

#3348 In huggingface/candle;

· lucasjinreal opened on Jan 30, 2026

Qwen3-ASR support !

Status: Open.

#3345 In huggingface/candle;

· lucasjinreal opened on Jan 30, 2026

Multiple versions of the same dependencies

Status: Open.

#3343 In huggingface/candle;

· bridiver opened on Jan 28, 2026

tensor upsample_cubic support

Status: Open.

#3339 In huggingface/candle;

· lucasjinreal opened on Jan 27, 2026

O_DIRECT when reading weight files

Status: Open.

#3337 In huggingface/candle;

· sempervictus opened on Jan 26, 2026

Consider using candle built-in interpolate op for image resize

Status: Open.

#3336 In huggingface/candle;

· lucasjinreal opened on Jan 26, 2026

DType enum changes in 0.9.2 patch release violated semver - please add #[non_exhaustive]

Status: Open.

#3333 In huggingface/candle;

· antimora opened on Jan 25, 2026