Repository navigation

All pull requests

Labels (11)Milestones (0)New pull request

Open

(400)

Closed

(2,438)

Fix the Metal sdpa causal mask when there are fewer queries than keys
#4062dhkts1 opened Oct 23, 2026
feat(core): mmap-backed GGUF loading for quantized tensors
#4061lu-zero opened Oct 23, 2026
Add a gelu_tanh op with f32-internal forward and backward kernels to candle-nn
#4060DrJesseGlass opened Oct 22, 2026Contributor
Add backward passes to the fused rms_norm, layer_norm and softmax_last_dim ops
#4059DrJesseGlass opened Oct 22, 2026Contributor
Keep the CPU RMSNorm scale in f32
#4057xgbtxy opened Oct 21, 2026
4
Metal: 64-token tile for the quantized matrix-matrix kernel
#4056dhkts1 opened Oct 21, 2026
Metal: QTensor::quantize_onto no longer leaves a zeroed buffer in the pool
#4055dhkts1 opened Oct 21, 2026
Metal quantized matmul: read the input from its own offset
#4054dhkts1 opened Oct 21, 2026
fix(kernels): only define the double atomicAdd shim in device code (CUDA 13)
#4053freezed-corpse-143 opened Oct 21, 2026
perf(cpu): use argminmax for argmin, argmax, min and max over the last dim
#4051jvdd opened Oct 20, 2026
fix(onnx): treat Shape end as exclusive and clamp start/end
#4050RaphaelFakhri opened Oct 20, 2026
Fix flip on non-contiguous tensors
#4048spicysharkco opened Oct 20, 2026
Metal: use 32-bit index math in strided kernels
#4045jakin010 opened Oct 18, 2026
Fix contiguous memory in Conv1D/2D and upgrade yanked zip dependency
#4044disruptivebarbie opened Oct 16, 2026
Fix ZIP 9 filename handling for NPZ and PTH loading
#4041Goldziher opened Oct 16, 2026
2
Add kev-rust: Kev-0.8B inference example (CPU and CUDA)
#4040fadhilelrizanda opened Oct 14, 2026
2
chore(deps): update zip requirement from 8.6.0 to 9.0.1
#4037dependabot[bot] opened Oct 13, 2026Bot
1
chore(deps): update minijinja requirement from 2 to 3
#4036dependabot[bot] opened Oct 13, 2026Bot
Fix Metal matmul on single-head and padded-stride views
#4033Cassian433 opened Oct 11, 2026
candle-kernels: fix int overflow in softmax, rmsnorm and layernorm
#4032lit0ne opened Oct 11, 2026
Fix Metal try_clone sharing the source buffer
#4030muzammil35 opened Oct 9, 2026
Fix double-paren links in quantized Qwen example READMEs
#4028pratikgx opened Oct 8, 2026
feat: support all LFM2.5 text models, including the 8B-A1B MoE
#4026ykhrustalev opened Oct 7, 2026
1
chore(deps): update cutile requirement from =0.3.1 to =0.4.0
#4025dependabot[bot] opened Oct 5, 2026Bot
Add signed step ranges to tensor indexing
#4022tianrking opened Oct 4, 2026

Navigation Menu

Repository navigation

All pull requests

Fix the Metal sdpa causal mask when there are fewer queries than keys

feat(core): mmap-backed GGUF loading for quantized tensors

Add a gelu_tanh op with f32-internal forward and backward kernels to candle-nn

Add backward passes to the fused rms_norm, layer_norm and softmax_last_dim ops

Keep the CPU RMSNorm scale in f32

Metal: 64-token tile for the quantized matrix-matrix kernel

Metal: `QTensor::quantize_onto` no longer leaves a zeroed buffer in the pool

Metal quantized matmul: read the input from its own offset

fix(kernels): only define the double atomicAdd shim in device code (CUDA 13)

perf(cpu): use argminmax for argmin, argmax, min and max over the last dim

fix(onnx): treat Shape end as exclusive and clamp start/end

Fix flip on non-contiguous tensors

Metal: use 32-bit index math in strided kernels

Fix contiguous memory in Conv1D/2D and upgrade yanked zip dependency

Fix ZIP 9 filename handling for NPZ and PTH loading

Add kev-rust: Kev-0.8B inference example (CPU and CUDA)

chore(deps): update zip requirement from 8.6.0 to 9.0.1

chore(deps): update minijinja requirement from 2 to 3

Fix Metal matmul on single-head and padded-stride views

candle-kernels: fix int overflow in softmax, rmsnorm and layernorm

Fix Metal try_clone sharing the source buffer

Fix double-paren links in quantized Qwen example READMEs

feat: support all LFM2.5 text models, including the 8B-A1B MoE

chore(deps): update cutile requirement from =0.3.1 to =0.4.0

Add signed step ranges to tensor indexing

Pull requests