Python package

python-package.yml

1064 workflow runs

Fix XPU quantize_4bit returning padded tensors that violate the op contract
requires action with the application:
Python package#3020: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:
Python package#3019: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:
Python package#3018: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:
Python package#3017: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:
Python package#3016: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:
Python package#3015: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:
Python package#3014: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:
Python package#3013: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:
Python package#3012: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:
Python package#3011: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:
Python package#3010:Pull request#2090openedbySIDDARTHAREDDY8
SIDDARTHAREDDY8:bhai-blitz-8-bitsandbytes-1
Sep 20, 2026Action required
[ROCm] Preserve float bits in SIMT DPP reduction
failed:
Python package#3009:Pull request#2089openedbyliminfei-amd
liminfei-amd:fix/2087-gfx1100-dpp-bitcast
Sep 15, 202630d 0h 5m 9s
fix(optim): reject bias_correction and adam_w_mode in LAMB
failed:
Python package#3008: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:
Python package#3007: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:
Python package#3006:Pull request#2055synchronizebySparshM8
SparshM8:fix/cpu-dequantize-1d-shape
Sep 10, 20261h 19m 18s
Fix Params4bit.to() leaving CPU-packed weights packed
failed:
Python package#3005:Pull request#2083openedby22elix3r
22elix3r:fix/params4bit-to-cpu-packing
Sep 10, 202630d 0h 0m 59s
Fix PytorchLARS updates when momentum is zero
failed:
Python package#3004:Pull request#2082openedbymottopanikeiku
mottopanikeiku:fix/issue-2079
Sep 10, 202630d 0h 5m 6s
Warn once when Linear4bit computes in float32 on GPU
failed:
Python package#3003: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:
Python package#3002: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:
Python package#3001:Pull request#2074openedbydependabot[bot]Bot
dependabot/github_actions/actions-6512b1d693
Sep 7, 202610m 35s
chore(deps): bump softprops/action-gh-release in the actions group
failed:
Python package#3000:Commit832d3c1pushedbydependabot[bot]Bot
dependabot/github_actions/actions-6512b1d693
Sep 7, 20269m 28s
[ROCm] Allow explicit GPU architecture override
failed:
Python package#2999: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:
Python package#2998:Pull request#2072synchronizebyrachitj27
rachitj27:feature/cpu-avx2-4bit-dequant
Sep 4, 202630d 0h 2m 54s
Add AVX2 path for CPU 4-bit (NF4/FP4) blockwise dequantization
failed:
Python package#2997:Pull request#2072openedbyrachitj27
rachitj27:feature/cpu-avx2-4bit-dequant
Sep 4, 202630d 0h 2m 54s
Fix CPU Hub kernel backend selection
completed successfully:
Python package#2996:Pull request#2065synchronizebySiluPanda
SiluPanda:fix/issue-2064
Sep 4, 202612m 24s

Navigation Menu

Python package

1064 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

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

chore(deps): bump softprops/action-gh-release 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 Python package

1064 workflow runs