Repository Issue Activity (beta)

#3582 Qwen3: causal_mask shaped (b, 1, tgt, tgt) from a b-independent buffer → wrong output for batch size > 1

#3580 Add the gpt-oss model (OpenAI open-weights MoE) to candle-transformers

#3515 candle-flash-attn vendored kernels ~13 months stale; long-context bf16 divergence vs PyTorch flash_attn 2.8.3

#3533 Unguarded allocation sites in candle-core/src/quantized/gguf_file.rs (DoS via crafted GGUF)

#3514 Contribute Qwen3.5/3.6 35B-A3B port + 3 fused perf kernels — packaging guidance?

#2168 No backward pass for `RmsNorm` if tensor is contiguous

#3569 candle-nn: ops::softmax_last_dim is apply_op1_no_bwd; severs autograd (same pattern as #2168 / PR #3526)

#3568 candle-nn: rotary_emb::rope is apply_op3_no_bwd; severs autograd (same pattern as #2168 / PR #3526)

#3567 Qwen2: prepare_attention_mask is bidirectional (no causal triangle) when attn_mask provided

#3566 MetalDevice::new panics on empty MTLCopyAllDevices() instead of returning Err

#1810 Vulkan support

#2877 Candle Inference ~8.5x Slower Than PyTorch on CPU

#3314 Is there any plan to support LFM 2.5 Audio , Text and Vision Models?

#3552 `panic!` when running on tvOS simulator with `mistral.rs`

#3541 16GB mem unable to build candle anymore

#3540 16GB mem unable to build candle anymore

#3534 Tensor::from_slice / from_raw_buffer return Ok when the declared shape does not match the storage element count

#3530 Candle leaves ~ model size of memory unused on Spark

#3389 conv1d with large groups (depthwise) is extremely slow on CUDA — O(groups) kernel launches

#3527 Cannot load tokenizer from gguf file

#2799 Siglip2 model support request

#2182 Whisper microphone example outputs gibberish

#2854 More examples about marian

#3497 The fact that Candle supports Metal is not mentioned in the mdbook (installation)

#3523 Phi4 mini multiple runs