Build PR Documentation

build_pr_documentation.yml

435 workflow runs

Fix XPU quantize_4bit returning padded tensors that violate the op contract
requires action with the application:
Build PR Documentation#1981:Pull request#2099openedbyjiqing-feng
jiqing-feng:xpu-fix-4bit-quant-shape
Oct 15, 2026Action required
[ROCm] Preserve float bits in SIMT DPP reduction
requires action with the application:
Build PR Documentation#1980:Pull request#2089synchronizebyliminfei-amd
liminfei-amd:fix/2087-gfx1100-dpp-bitcast
Oct 12, 2026Action required
Optimize NF4 single-token GEMV on Blackwell
requires action with the application:
Build PR Documentation#1979:Pull request#2098synchronizebyPiotrPotocki
PiotrPotocki:optimind/nf4-fused-gemv-setup
Oct 10, 2026Action required
Optimize NF4 single-token GEMV on Blackwell
requires action with the application:
Build PR Documentation#1978:Pull request#2098openedbyPiotrPotocki
PiotrPotocki:optimind/nf4-fused-gemv-setup
Oct 8, 2026Action required
Support skip_zeros in the XPU Triton optimizer kernels
requires action with the application:
Build PR Documentation#1977:Pull request#2097synchronizebyjiqing-feng
jiqing-feng:xpu-triton-skip-zeros
Oct 8, 2026Action required
Support skip_zeros in the XPU Triton optimizer kernels
requires action with the application:
Build PR Documentation#1976:Pull request#2097openedbyjiqing-feng
jiqing-feng:xpu-triton-skip-zeros
Oct 8, 2026Action required
fix: wrong absmax indexing in AVX512 dequantizeBlockwise4bitCpu
requires action with the application:
Build PR Documentation#1975:Pull request#2096openedbyNanoCode012
NanoCode012:patch-1
Oct 7, 2026Action required
MPS: only use the Hub GEMV kernel when K is a multiple of blocksize
requires action with the application:
Build PR Documentation#1974:Pull request#2094openedbyegeozkoc
egeozkoc:fix/mps-hub-gemv-unaligned-k
Sep 30, 2026Action required
test: align post-reload max_error_count with pre-reload check in test_optimizer32bit
requires action with the application:
Build PR Documentation#1973:Pull request#2093openedbyCS-liujf
CS-liujf:fix/lion-reload-threshold
Sep 23, 2026Action required
Fix device parametrization for optimizer benchmarks
requires action with the application:
Build PR Documentation#1972:Pull request#2092openedbyemme1t
emme1t:fix/optimizer-benchmark-device
Sep 22, 2026Action required
fix: pass backend="cpu" when requesting CPU gemm_4bit_forward kernel
requires action with the application:
Build PR Documentation#1971:Pull request#2090openedbySIDDARTHAREDDY8
SIDDARTHAREDDY8:bhai-blitz-8-bitsandbytes-1
Sep 20, 2026Action required
[ROCm] Preserve float bits in SIMT DPP reduction
requires action with the application:
Build PR Documentation#1970:Pull request#2089openedbyliminfei-amd
liminfei-amd:fix/2087-gfx1100-dpp-bitcast
Sep 15, 2026Action required
fix(optim): reject bias_correction and adam_w_mode in LAMB
failed:
Build PR Documentation#1969:Pull request#2088openedbyConnorMoss02
ConnorMoss02:fix/lamb-reject-unsupported-args
Sep 15, 202630d 0h 1m 40s
fix(optim): reject nesterov and dampening in SGD and LARS
failed:
Build PR Documentation#1968:Pull request#2085openedbyConnorMoss02
ConnorMoss02:fix/sgd-lars-reject-unsupported-nesterov-dampening
Sep 11, 202630d 0h 3m 10s
fix(cpu): correctly handle 1D output shape in dequantize_4bit
failed:
Build PR Documentation#1967:Pull request#2055synchronizebySparshM8
SparshM8:fix/cpu-dequantize-1d-shape
Sep 10, 20261h 19m 17s
Fix Params4bit.to() leaving CPU-packed weights packed
failed:
Build PR Documentation#1966:Pull request#2083openedby22elix3r
22elix3r:fix/params4bit-to-cpu-packing
Sep 10, 202630d 0h 0m 56s
Fix PytorchLARS updates when momentum is zero
failed:
Build PR Documentation#1965:Pull request#2082openedbymottopanikeiku
mottopanikeiku:fix/issue-2079
Sep 10, 202630d 0h 5m 5s
[WIP] Prototype original-parameter AdamW8bit state for FSDP1
failed:
Build PR Documentation#1964:Pull request#2081openedbygss10282025
gss10282025:prototype-fsdp-original-parameter-state
Sep 10, 202630d 0h 5m 25s
Warn once when Linear4bit computes in float32 on GPU
failed:
Build PR Documentation#1963:Pull request#2077openedbythangdangjp
thangdangjp:feature/warn-fp32-compute-dtype
Sep 9, 202630d 0h 2m 30s
Make AdEMAMix32bit inherit AdEMAMix so it allocates the double state buffer
failed:
Build PR Documentation#1962:Pull request#2076openedbycaiotheodoro
caiotheodoro:ademamix32bit-double-buffer
Sep 8, 202630d 0h 3m 10s
chore(deps): bump softprops/action-gh-release from 3.0.2 to 3.0.3 in the actions group
completed successfully:
Build PR Documentation#1961:Pull request#2074openedbydependabot[bot]Bot
dependabot/github_actions/actions-6512b1d693
Sep 7, 20261m 7s
[ROCm] Allow explicit GPU architecture override
failed:
Build PR Documentation#1960:Pull request#2073openedbyLOGO127
LOGO127:fix/rocm-arch-env-override
Sep 4, 202630d 0h 2m 55s
Add AVX2 path for CPU 4-bit (NF4/FP4) blockwise dequantization
failed:
Build PR Documentation#1959:Pull request#2072synchronizebyrachitj27
rachitj27:feature/cpu-avx2-4bit-dequant
Sep 4, 202630d 0h 2m 53s
Add AVX2 path for CPU 4-bit (NF4/FP4) blockwise dequantization
failed:
Build PR Documentation#1958:Pull request#2072openedbyrachitj27
rachitj27:feature/cpu-avx2-4bit-dequant
Sep 4, 202630d 0h 2m 53s
Fix CPU Hub kernel backend selection
completed successfully:
Build PR Documentation#1957:Pull request#2065synchronizebySiluPanda
SiluPanda:fix/issue-2064
Sep 4, 20261m 15s

Navigation Menu

Build PR Documentation

435 workflow runs

Fix XPU `quantize_4bit` returning padded tensors that violate the op contract

[ROCm] Preserve float bits in SIMT DPP reduction

Optimize NF4 single-token GEMV on Blackwell

Optimize NF4 single-token GEMV on Blackwell

Support `skip_zeros` in the XPU Triton optimizer kernels

Support `skip_zeros` in the XPU Triton optimizer kernels

fix: wrong absmax indexing in AVX512 dequantizeBlockwise4bitCpu

MPS: only use the Hub GEMV kernel when K is a multiple of blocksize

test: align post-reload max_error_count with pre-reload check in test_optimizer32bit

Fix device parametrization for optimizer benchmarks

fix: pass backend="cpu" when requesting CPU gemm_4bit_forward kernel

[ROCm] Preserve float bits in SIMT DPP reduction

fix(optim): reject bias_correction and adam_w_mode in LAMB

fix(optim): reject nesterov and dampening in SGD and LARS

fix(cpu): correctly handle 1D output shape in dequantize_4bit

Fix Params4bit.to() leaving CPU-packed weights packed

Fix PytorchLARS updates when momentum is zero

[WIP] Prototype original-parameter AdamW8bit state for FSDP1

Warn once when Linear4bit computes in float32 on GPU

Make AdEMAMix32bit inherit AdEMAMix so it allocates the double state buffer

chore(deps): bump softprops/action-gh-release from 3.0.2 to 3.0.3 in the actions group

[ROCm] Allow explicit GPU architecture override

Add AVX2 path for CPU 4-bit (NF4/FP4) blockwise dequantization

Add AVX2 path for CPU 4-bit (NF4/FP4) blockwise dequantization

Fix CPU Hub kernel backend selection

Workflow runs for Build PR Documentation

435 workflow runs