[pull] main from huggingface:main - #41

Open

pull[bot] wants to merge 456 commits into

EricLBuehler:mainfrom

huggingface:main

This pull request is big! We're only showing the most recent 250 commits

Commits on Sep 8, 2025

Merge pull request #3077 from zhanluxianshen/typo-candle-core
ivarflakstad
authored
87fadf6
Fix metal exports (#3081)
ivarflakstad
authored
0950959

Commits on Sep 9, 2025

Merge branch 'main' into metal-tensor-fix-send-sync
ivarflakstad
committed
a7fbc63
Merge pull request #3079 from huggingface/metal-tensor-fix-send-sync
ivarflakstad
authored
65055f6
Merge pull request #3062 from davenpi/fix/core-basics-example
ivarflakstad
authored
b1dbce0
Add CUDA 13 support (#3078)
jfernandez
authored
8045af9
Fix indentation
ivarflakstad
authored
97594d2
Fix indentation (ok but for real)
ivarflakstad
authored
038e28b
Merge pull request #2937 from ljt019/fix-phi3-kv-cache-reset
ivarflakstad
authored
372c9cf

Commits on Sep 12, 2025

add impl for mish activation function (#3051)
oa-root
authored
41a674c

Commits on Sep 18, 2025

Upgrade ug dep for CUDA 13 support
grahamking
committed
dd12467

Commits on Sep 20, 2025

Merge pull request #3089 from grahamking/main
ivarflakstad
authored
1a699fb

Commits on Sep 22, 2025

Commits on Sep 30, 2025

Commits on Oct 1, 2025

Commits on Oct 2, 2025

Commits on Oct 4, 2025

Commits on Oct 5, 2025

Skip unsupported quantized matmul tests for metal (#3115)
ivarflakstad
authored
671de1d

Commits on Oct 6, 2025

Fix beit on metal by adding additional affine implementations (#3116)
ivarflakstad
authored
bcc34bc

Commits on Oct 7, 2025

Commits on Oct 8, 2025

Commits on Oct 9, 2025

Fix failing `cuda` build (#3121)
LLukas22
authored
7b8f2b4
feat: add metal_if_available method for graceful Metal fallback (#3041)
xavierforge
authored
cc967fc
Fix metal quantized to_float calls (#3123)
ivarflakstad
authored
bffa5e1

Commits on Oct 13, 2025

Commits on Oct 16, 2025

Update modernbert.rs (#3010)
whitebox2
authored
f601fd8
Update dependencies (#3135)
ivarflakstad
authored
701205a

Commits on Oct 17, 2025

Ensure output of Transpose is contiguous to prevent downstream MatMul from crashing (#3088)
kshitijl
authored
1febb7b

Commits on Oct 18, 2025

In the BERT example: apply the attention mask from tokenization during pooling (#3085)
kshitijl
authored
2bce4e5

Commits on Oct 20, 2025

Skip q8k and q8_1 tests on cuda (#3140)
ivarflakstad
authored
a52f22f

Commits on Oct 23, 2025

Implement qwen3 vl
EricLBuehler
committed
36b7517
Clippy
EricLBuehler
committed
fd379c5
Bump candle version to 0.9.2-alpha.1 (#3146)
ivarflakstad
authored
59aeed4
Remove unused
EricLBuehler
committed
5b7858c
Add Qwen 3 VL to candle-transformers
EricLBuehler
authored
e3228c1
Improve candle example buildtime downloader (#3147)

ivarflakstad
and
matthewhaynesonline
authored
d312da2

Commits on Oct 25, 2025

Commits on Oct 28, 2025

fix: build errors for compute cap 7.5 (#3142)

neksodebe
and
ivarflakstad
authored
fab0c45
Update cargo build instructions to use double colon syntax (#3132)

matthewhaynesonline
and
ivarflakstad
authored
a05b549
Add flash attn v3: `candle-flash-attn-v3` (#3152)
EricLBuehler
authored
8f27f5c

Commits on Oct 30, 2025

Add nccl feature to candle-core (#3155)
EricLBuehler
authored
7669ed1

Commits on Oct 31, 2025

Commits on Nov 1, 2025

Add unsafe from_storage apis (#3156)
EricLBuehler
authored
d4545eb

Commits on Nov 6, 2025

[Metal] Ensure metal backend is send/sync via status semaphore (#3164)
ivarflakstad
authored
b06a02c

Commits on Nov 7, 2025

Add sqrt2 as constant for gelu_erf and use `libm` erf (#3168)
vrdn-23
authored
ade0918

Commits on Nov 8, 2025

Commits on Nov 9, 2025

Use Tensor::argmax instead of manual cpu impl (#3173)
ivarflakstad
authored
bf3d3f2

Commits on Nov 11, 2025

Commits on Nov 14, 2025

feat(candle-nn) ConcatKvCache for 2-5x GPU speedup on autoregressive generation (#3143)

DrJesseGlass
and
ivarflakstad
authored
60252cc

Commits on Nov 17, 2025

Add `cublas_handle` api, update safetensors (#3192)
EricLBuehler
authored
8ebfc22
Update CI (#3194)
ivarflakstad
authored
ab56dfe

Commits on Nov 18, 2025

Commits on Nov 20, 2025

Commits on Nov 25, 2025

Add lld installation and test steps for Linux (#3213)
haricot
authored
b801ef6
Add dummy dtypes (#3195)

EricLBuehler
and
ivarflakstad
authored
01bea21
Add more misc. changes from candle fork (#3196)

authored
95ea453

Commits on Nov 30, 2025

.gitignore: add .zed to ignored editor configs (#3218)
slckl
authored
2ac3fe0

Commits on Dec 2, 2025

chore(dep): bump cudarc to 0.18.1 (#3219)
mayocream
authored
c39d5f0

Commits on Dec 3, 2025

Hotfix: Bump float8 to 0.5.0 (#3223)
EricLBuehler
authored
08d7b64

Commits on Dec 4, 2025

[Metal] Make fast math mode optional (#3205)
ivarflakstad
authored
2664a21
Update pyo3 (#3202)
ivarflakstad
authored
9ede204

Commits on Dec 6, 2025

[Metal] unary and affine improvements (#3230)
ivarflakstad
authored
3d3cc49

Commits on Dec 8, 2025

[Metal] binary improvements (#3231)
ivarflakstad
authored
72238a7
fix(metal): add missing softcapping field to AttnParams struct (#3233)
amritsingh183
authored
d91be02
Format sdpa (#3235)
EricLBuehler
authored
2a797ea

Commits on Dec 9, 2025

Fix metal argmax (#3238)
EricLBuehler
authored
d23664f

Commits on Dec 10, 2025

[Metal] further improve unary and binary (#3239)
ivarflakstad
authored
73fd9c3
[Metal] cast improvements (#3241)
ivarflakstad
authored
e33d776

Commits on Dec 14, 2025

[Metal] Improve ternary further (#3242)
ivarflakstad
authored
4b46187

Commits on Dec 16, 2025

Bump candle version to 0.9.2-alpha.2 (#3248)
ivarflakstad
authored
8839457

Commits on Dec 21, 2025

add candle flash attention 3 copyright markers (#3256)
michaelfeil
authored
689d255

Commits on Dec 23, 2025

fix: replace deprecated cudarc memcpy methods (#3228)

DrJesseGlass
and
ivarflakstad
authored
ab6d97e
Support Fused MoE & Qwen3 GGUF MoE models (#3221)

guoqingbao
and
ivarflakstad
authored
f2d5aab

Commits on Dec 24, 2025

Upgrade GitHub Actions for Node 24 compatibility (#3255)
salmanmkc
authored
049c06d

Commits on Dec 26, 2025

Commits on Dec 27, 2025

Fix remnant memcpy_stod call (#3267)
ivarflakstad
authored
63437a4

Commits on Dec 30, 2025

Commits on Dec 31, 2025

add HuberLoss (#3252)

donjuanplatinum
and
ivarflakstad
authored
5de3d0f
feat!: Make `ug` dependency optional (#3268)

DanikVitek
and
ivarflakstad
authored
d8fb848

Commits on Jan 2, 2026

Add Z-Image Text-to-Image Generation Support (#3261)

SpenserCai
and
ivarflakstad
authored
3a0d1cb

Commits on Jan 3, 2026

fix(candle-kernels): conditionally link stdc++ for non-MSVC targets (#3278)
Elvis339
authored
43be23c

Commits on Jan 4, 2026

replace cutlass submodule references with explicit build step (#3234)

jacobgorm
and
ivarflakstad
authored
a4ad7c7

Commits on Jan 6, 2026

Commits on Jan 7, 2026

Mamba2 implementation (#3264)
Anri-Lombard
authored
42a4edc
feat: paddleocr-vl model and example (#3273)
danielclough
authored
f526033

Commits on Jan 14, 2026

chore(dep): bump cudarc to 0.18.2 (#3293)
staceymelville
authored
f2d30fb
Add SmolLM3: Full and Quantized Implementation (#3180)

DrJesseGlass
and
ivarflakstad
authored
dbb8c2d
example: add quantized qwen3 wasm with SIMD optimizations (#3159)

DrJesseGlass
and
ivarflakstad
authored
a2029da

Commits on Jan 17, 2026

Hotfix: Remove fastmath from candle-kernels (#3309)
EricLBuehler
authored
aaf5c86
feat: add quantized lfm2 model support (#3244)

fffonion
and
ivarflakstad
authored
261f727
Support new arch of GLM4 GGUF models (#2992)

guoqingbao
and
ivarflakstad
authored
23182cf

Commits on Jan 18, 2026

feat: simplify metal reduce kernels and standardize on u32 indexing (#3285)

drbh
and
ivarflakstad
authored
cc8ec5e

Commits on Jan 19, 2026

Commits on Jan 21, 2026

Metal GEMM Dynamic Tile Selection and Batch Collapse Optimization (#3313)

SpenserCai
and
ivarflakstad
authored
06cb713

Commits on Jan 22, 2026

Update deps (#3320)

ivarflakstad
and
EricLBuehler
authored
8d5873b

Commits on Jan 23, 2026

CUDA Tensor::toDevice copies directly via memcopy_dtod (#3312)

krampenschiesser
and
EricLBuehler
authored
88ed791

Commits on Jan 24, 2026

[Cuda] Use upstream bindgen_cuda crate (#3328)
ivarflakstad
authored
f041b87
Bump candle version to 0.9.2 (#3329)
ivarflakstad
authored
e53310d
Add dep versioning for candle-flash-attn-build (#3330)
ivarflakstad
authored
3b39794

Commits on Feb 4, 2026

Commits on Feb 10, 2026

Use cudaforge for kernel build (#3346)

guoqingbao
and
EricLBuehler
authored
c3bb5bf
Add candle-video library for text-to-video generation in README.md (#3365)
FerrisMind
authored
f2cb5b4

Commits on Feb 15, 2026

fix: disable bf16 WMMA kernels on pre-Ampere GPUs (compute cap < 80) (#3367)
asglover
authored
a0dbd8b

Commits on Feb 17, 2026

feat: rwkv_v7 model and examples (#3372)
danielclough
authored
2bd0256

Commits on Mar 22, 2026

fix: conv2d_tiled produces wrong results when C == H == W (#3405)
developer0hye
authored
bf9e950

Commits on Mar 28, 2026

Commits on Mar 29, 2026

feat: add eps() and remove_mean() getters to LayerNorm and RmsNorm (#3387)

jgremes
and
ivarflakstad
authored
f7e39a4
[Metal] Use StorageModePrivate for intermediate compute buffers (#3416)

jamisonl
and
ivarflakstad
authored
e0e33e9
fix: exclude tokenizers dependency on wasm32 targets (#3414)

ilnmtlbnm
and
ivarflakstad
authored
6b4d8a1
Guard against NULL architecture pointer on simulators (#3392)

setoelkahfi
and
ivarflakstad
authored
c6a4649
fix(metal): buffer pool rounds 2 down by 1 (#3394)

setoelkahfi
and
ivarflakstad
authored
a7d8427

Commits on Mar 31, 2026

Bump candle version to 0.10.0 (#3432)
ivarflakstad
authored
153cadd

Commits on Apr 2, 2026

Commits on Apr 8, 2026

Merge pull request #3442 from huggingface/security/pin-actions-to-sha
paulinebm
authored
c42e1fe

Commits on Apr 9, 2026

Handle case with gemma4 without safetensors index (#3457)
EricLBuehler
authored
34625ab

Commits on Apr 13, 2026

Add fast CUDA MMVQ GGUF kernels (#3463)
EricLBuehler
authored
b503458

Commits on Apr 15, 2026

Commits on Apr 16, 2026

Fix metal GgmlDType::BF16 mapping (#3471)
ivarflakstad
authored
5d20d1f
Fix fmt and clippy remnants (#3472)
ivarflakstad
authored
cce7901

Commits on Apr 22, 2026

Fix: use MTLCopyAllDevices() for reliable Metal device enumeration (#3449)

romnn
and
ivarflakstad
authored
c5d7d49
fix clippy lints surfaced by Rust 1.95 (#3481)
tomsanbear
authored
bb2d400
update `erf::polynomial` (#1413)

chris-ha458
and
ivarflakstad
authored
c8c7663
Add rustup wasm doc to wasm example (#1438)

jk2K
and
ivarflakstad
authored
2bbbb4e
Extend `GradStore` public functionality (#1483)

agerasev
and
ivarflakstad
authored
9e71760
Remove unnecessary task (#1925)
kejcao
authored
7fa19d2
fix: candle-book paths (#3386)

authored
b43326e
Remove unwrap()s from candle-metal-kernels/src/metal/device.rs (#3382)
jacobgorm
authored
5bd5618

Commits on Apr 23, 2026

Optimization for CPU Causal Flash Attention (integrated into Qwen3) (#3254)

authored
5447a87

Commits on May 14, 2026

[Metal] Concurrent dispatching (#3511)
ivarflakstad
authored
b7eb2f0

Commits on May 15, 2026

[Metal] Improved inter-encoder sync and gemv (#3532)
ivarflakstad
authored
ad4298f

Commits on May 16, 2026

Commits on May 24, 2026

Fix FAv3 packing and symbol overlap (#3559)
EricLBuehler
authored
6a05b4b
Add htod cache for cuda graphs (#3564)
EricLBuehler
authored
cabbc30

Commits on May 26, 2026

Improve CUDA stream consistency (#3565)
EricLBuehler
authored
3ba428d
fix(kernels): provide __hmax_nan/__hmin_nan fallback for sm_75 (Turing) (#3558)

4mGLn
and
4aMgLn4
authored
1d7e927

Commits on May 28, 2026

chore(deps): update symphonia requirement from 0.5.3 to 0.6.0 (#3546)

dependabot[bot]
and
ivarflakstad
authored
dcc708e
chore(deps): update hf-hub requirement from 0.4.1 to 0.5.0 (#3545)

dependabot[bot]
and
ivarflakstad
authored
14f923b
chore(deps): update rubato requirement from 1 to 2 (#3544)

dependabot[bot]
and
ivarflakstad
authored
d2ba757

Commits on May 29, 2026

Binary broadcast scalar support (#3487)
ivarflakstad
authored
09b9145

Commits on Jun 1, 2026

metal: add copy2d kernels for I16 / I32 (#3478)

tomsanbear
and
ivarflakstad
authored
5404348
Improved neon vec dot (#3570)
ivarflakstad
authored
39355c6

Commits on Jun 4, 2026

metal: fix RMSNorm NaN on large-magnitude F32 inputs (#3477)
tomsanbear
authored
1a63038

Commits on Jun 6, 2026

Cap allocations in the GGUF loader (#3556)

HueCodes
and
ivarflakstad
authored
c72a17f
chore(deps): update enterpolation requirement from 0.2.1 to 0.3.0 (#3574)

dependabot[bot]
and
ivarflakstad
authored
b5a101a

Commits on Jun 7, 2026

Commits on Jun 8, 2026

Add LFM2.5 (Liquid Foundation Model 2.5) support (#3400)

Jacqkues
and
claude
authored
3d3d9c4

Commits on Jun 17, 2026

Commits on Jun 18, 2026

Commits on Jun 19, 2026

Commits on Jun 20, 2026

Remove flash attn specific rayon threadpool. Use device.with_context around entire fwd pass instead to access the shared threadpool (#3634)
ivarflakstad
authored
29a15c2

Commits on Jun 23, 2026

Support cuda 13.3 (#3642)
EricLBuehler
authored
a9667ca

Commits on Jun 24, 2026

Parameter cache for CUDA kernel launch (#3598)
guoqingbao
authored
f53ed3b

Commits on Jun 25, 2026

Support embedding forward pass for ggml quants (#3644)
EricLBuehler
authored
5152ef6
[CPU] Quantized perf improvements (#3643)
ivarflakstad
authored
c63cb20
Add paged flash-attn kernels (#3655)
EricLBuehler
authored
9bcfd98

Commits on Jun 26, 2026

Bump candle version to 0.11.0 (#3658)
ivarflakstad
authored
31f35b1

Commits on Jul 23, 2026

Fix clippy warnings for Rust 1.97 (#3754)
GregoryBolshakov
authored
35e1bd7
Fix Metal pipeline cache source identity (#3760)

tomsanbear
and
ivarflakstad
authored
2518855

Commits on Jul 27, 2026

Fix duplicated words in documentation (#3786)
cupkk
authored
5e9b2d6
feat(metal): register external buffers in the device residency set (#3771)

0rvar
and
ivarflakstad
authored
578ed5b
[CPU] Fix wrong offset for non-contiguous conv2d im2col kernels (#3736)

Sammy-Dabbas
and
ivarflakstad
authored
21cca0b

Commits on Jul 30, 2026

fix(qwen3): build causal mask batch-independently (#3582) (#3586)
pjdurden
authored
6e823a4

Commits on Aug 5, 2026

Commits on Aug 6, 2026

Use shared locks for Metal cache hits (#3761)
tomsanbear
authored
6f74e7c

Commits on Aug 13, 2026

fix(models): centralize the additive causal mask, build it batch-independently (#3879)
pjdurden
authored
3b7af4a
fix(cpu): avoid very recently stable aarch64 fp16 vector type (#3845)

jamesbrink
and
ivarflakstad
authored
7b59bb9
Fix concrete shape element count validation (#3813)

ccyyy1023
and
ivarflakstad
authored
94c83c4
Fix the ug+metal build against the guard-based encoder API (#3855)

tomsanbear
and
ivarflakstad
authored
16d47ab
Fix Metal rank-5+ strided reduce reading out of bounds (#3873)

waamm
and
ivarflakstad
authored
2395577
Fix repeat with zero factors (#3877)
charpty
authored
defe4bb
Fix CPU argsort for offset tensor views (#3875)
LegenDairy93
authored
361034c
Avoid copying a rank-2 rhs in broadcast_matmul (#3872)
pjmalandrino
authored
bd4cdad

Commits on Aug 14, 2026

fix(metal): surface command buffer errors that occur during execution (#3862)

oetiker
and
ivarflakstad
authored
091395a

Commits on Aug 15, 2026

Fix FlashAttention v3 null tile count semaphore, stream, and LSE sizing (#3892)
EricLBuehler
authored
2a13b0f