Repository navigation

All pull requests

Labels (47)Milestones (2)New pull request

Open

(77)

Closed

(718)

Invalidate CPU quantization LUTs for complete codebook changes
#2141Afloat16 opened Nov 13, 2026
Respect metadata strides in CPU blockwise quantization
#2140Afloat16 opened Nov 13, 2026
Support saving and restoring quantized embedding checkpoints
#2139Afloat16 opened Nov 13, 2026
Preserve INT8 row scales when parameter conversions omit a device
#2138Afloat16 opened Nov 13, 2026
Keep quantized embedding output dtype in sync with module conversions
#2137Afloat16 opened Nov 13, 2026
Reload quantized Linear8bitLt checkpoints after the first forward
#2136Afloat16 opened Nov 12, 2026
Honor optimizer checkpoint load hooks for state migration
#2135Afloat16 opened Nov 12, 2026
Preserve module ownership and shared state in Params4bit deepcopy
#2134Afloat16 opened Nov 12, 2026
Forward StableEmbedding device and dtype as factory keywords
#2133Afloat16 opened Nov 12, 2026
Use the normalization parameter dtype in StableEmbedding
#2132Afloat16 opened Nov 12, 2026
Handle non-contiguous inputs in LLM.int8() CUDA ops
#2131Mazukiri opened Nov 12, 20267 tasks done
Move cached int8 linear weights when moving parent modules
#2129tgghzyh opened Nov 12, 2026
Avoid copying the full scale table in 4-bit embedding lookups
#2128tgghzyh opened Nov 11, 2026
perf: avoid repeated INT8 cast-warning records
#2127yfukuda-oqc opened Nov 11, 2026
docs: fix the LICENSE badge link
#2125cloudgpuapp opened Nov 10, 2026
fix(gaudi): default the plugin version when metadata has no Version
#2124devinhuang2001 opened Nov 9, 2026
test(mps): verify 4-bit dequantization values and Hub dispatch
#2123egeozkoc opened Nov 8, 2026Contributor
1
chore(deps): bump the actions group across 1 directory with 5 updates
#2120dependabot[bot] opened Nov 2, 2026Bot
1
Fix CPU quantize_blockwise to pick the nearest code
#2119damienen opened Nov 2, 2026
Do not freeze zero-norm parameters in LAMB/LARS trust-ratio clipping
#2117Andreas-Hum opened Nov 1, 2026
Fix INT8 weight-gradient quantization and CUDA operand layout
#2115devinhuang2001 opened Oct 30, 2026
Mock torch.version.hip in CUDA override tests
#2114kudomcho opened Oct 29, 20262 of 3 tasks
1
Fix fp32 4-bit GEMV test tolerance for ROCm/gfx1200
#2113kudomcho opened Oct 29, 20263 of 4 tasks
1
Fix k-bit bias gradients for non-2D inputs
#2111anh-nguyenn opened Oct 26, 2026
Fix MatMul8bitFp backward for non-2D inputs
#2109anh-nguyenn opened Oct 26, 2026

Pull requests