Go
Home
Pricing
FAQ
Install
Home
Pricing
FAQ
Install
Login
via GitHub
intel/auto-round
Pull Requests
Commits
Open
Closed
fix: fix DeepSeek V4 model-free NVFP4 E5M3 layer overrides
#2658 opened 2026-11-19 13:39 by
changwangss
fix(nvfp4): prevent invalid global scales in MoE exports
#2657 opened 2026-11-19 12:42 by
lkk12014402
0.17.0
[model-free] Split fused-expert block-FP8 *_scale_inv and record real index total_size
#2656 opened 2026-11-19 08:27 by
XuehaoSun
fix(ark): support XPU graph capture for MoE GEMM
#2655 opened 2026-11-19 07:42 by
Zhenzhong1
feat: support TR-HASH moe for tuned and model-free quantization
#2654 opened 2026-11-19 05:32 by
n1ck-guo
fix: update format handling in quantization configuration for model-free flow
#2653 opened 2026-11-19 02:01 by
xin3he
fix: don't substring-match org/name calibration datasets
#2651 opened 2026-11-18 15:51 by
0z5a
fix: enable DeepSeek V4 native FP4 requantization and preserve export names
#2649 opened 2026-11-18 13:22 by
changwangss
feat: introduce AR_NVFP_ACT_AMAX_ALPHA for scaling NVFP activation amax during calibration
#2647 opened 2026-11-18 07:34 by
xin3he
fix: apply the stored zero-point offset in the zp fused kernels
#2646 opened 2026-11-18 04:21 by
lxcxjxhx
perf: run QuantLinearFunction on the fused triton kernels
#2645 opened 2026-11-18 04:18 by
lxcxjxhx
fix: mask the packed tail in the transposed fused triton matmul
#2644 opened 2026-11-18 04:16 by
lxcxjxhx
test: check the fused triton matmul against a plain-torch reference
#2643 opened 2026-11-18 04:08 by
lxcxjxhx
fix: mask the K tail in the fused triton matmul
#2642 opened 2026-11-18 04:03 by
lxcxjxhx
fix: wrap GPTQ's -1 group marker in the fused triton matmul
#2641 opened 2026-11-18 04:00 by
lxcxjxhx
test: pin the two triton int2/4/8 matmul paths to the same result
#2640 opened 2026-11-18 03:58 by
lxcxjxhx
fix: keep the triton autotuner from raising when two configs tie
#2639 opened 2026-11-18 03:54 by
lxcxjxhx
docs: align two data-type docstrings with the signatures they document
#2638 opened 2026-11-18 03:30 by
lxcxjxhx
Support LLMC Format for Exported HMT Models
#2629 opened 2026-11-15 13:06 by
lkk12014402
fix(calib): support dataset config specs and recover from streaming schema errors (#2616)
#2628 opened 2026-11-15 04:10 by
Prajwal7-2002
fix(rrq): address PR #2308 review feedback
#2626 opened 2026-11-13 10:29 by
luoyu-intel
feat(examples): add NVFP4 quantization recipe for dots.mocr
#2625 opened 2026-11-13 09:48 by
Jagadeeshwaran-J
auto-scheme exp backup
draft
won't merge
#2613 opened 2026-11-10 09:11 by
WeiweiZhang1
add hessian for auto scheme
draft
WIP
#2612 opened 2026-11-10 07:33 by
n1ck-guo
Update calibration_acc.md
#2611 opened 2026-11-10 06:36 by
ZaneMark
refine autoscheme algorithm
#2609 opened 2026-11-10 05:18 by
wenhuach21
docs: add bilingual MXFP channel transform research
#2590 opened 2026-11-06 02:23 by
a32543254
Add fuzz tests for parsing and path safety in auto_round
#2577 opened 2026-11-04 00:32 by
chensuyue
feat: shard the AWQ pipeline under --data_parallel
#2575 opened 2026-11-03 21:48 by
avtc
Refactor quantization functions and enhance backend compatibility checks
#2562 opened 2026-10-30 14:26 by
chensuyue
Older