intel/auto-round

Pull Requests Commits

Refactor delta_loss and utils modules: add _resolve_dp_max_states function for DP beam width selection; enhance logging with _log_scheme_loss_matrix and _build_layer_config_header_rows; remove _apply_head_trick tests.

xin3he committed 44 days ago

426217f6

Refactor utils and delta_loss modules to enhance logging and expert score handling; streamline quantization scheme representation and layer configuration logging.

xin3he committed 45 days ago

9780d177

Enhance documentation and comments for AutoScheme scoring parameters and methods; clarify default values for nsamples in environments.

xin3he committed 45 days ago

1c5c35c4

Refactor _gen_layer_config to simplify nsamples and seqlen assignment for MoE models

xin3he committed 45 days ago

56316a03

Add environment variables for AutoScheme nsamples and batch size; update documentation and tests

xin3he committed 45 days ago

f2c6c230

revert nsamples change

xin3he committed 45 days ago

7b13fd4a

Enhance model_forward_low_gpu to support scheme tagging and improve logging of batch average loss

xin3he committed 45 days ago

0a78c646

set lm_head grad_mode=True

xin3he committed 46 days ago

115b5e49

add shared loss for inactived experts and refine log

xin3he committed 46 days ago

6b2d5a59

update log in autoscheme

xin3he committed 46 days ago

67df125c

fix bug of autoscheme

xin3he committed 46 days ago

6c564fba

fix auto_scheme gguf format wrong bits if use tiny model (#1953)

n1ck-guo committed 47 days ago

Verified b1a6baf9

speed up awq ci (#1954)

WeiweiZhang1 committed 50 days ago

Verified 9468e52c

add other non-integer scheme support for awq (#1892)

WeiweiZhang1 committed 50 days ago

Verified 5a3ade8c

Remove kernels version limit in test (#1945)

chensuyue committed 51 days ago

Verified 142e91b4

Use shared tokenizer save helper in exporters (#1951)

changwangss committed 51 days ago

Verified 436b9db6

update ARK README (#1922)

luoyu-intel committed 51 days ago

Verified f42a0421

Implement shard parallelism configuration and update documentation (#1942)

xin3he committed 52 days ago

Verified 7e2b69d4

fix saving hadamard rotation model without rotation config. (#1933)

lkk12014402 committed 52 days ago

Verified 6a2e5ad3

fix few gguf bugs of hf-name and checkpoint name (#1939)

n1ck-guo committed 52 days ago

Verified 126793d5

Add flashinfer-jit-cache installation to cuda unit test setup (#1947)

XuehaoSun committed 53 days ago

Verified 349b0418

fix issue of step3p7 and CI failure (#1929)

xin3he committed 53 days ago

Verified 1c4baa03

Update PyTorch and torchvision versions to 2.12.1 and 0.27.1 in CI (#1943)

XuehaoSun committed 53 days ago

Verified 354c6a3b

Update CPU agent pool (#1936)

chensuyue committed 53 days ago

Verified 35650a09

Fix SIGSEGV in pack-weight unpack/GEMM on malformed blob (#1928)

luoyu-intel committed 54 days ago

Verified 495e2539

Refine AutoScheme logic for gguf, reduce memory consumption and improve speed (#1916)

wenhuach21 committed 54 days ago

Verified 0c9c5b1d

DDP for multiple cards (#1897)

yiliu30 committed 56 days ago

Verified 33b7df0e

reorder UT for CUDA CI memory leak (#1938)

xin3he committed 58 days ago

Verified 0cc47c69

update what's new: vllm-omni integrates AutoROund (#1931)

xin3he committed 58 days ago

Verified 759a9ad7

Enhance b70 check (#1937)

yiliu30 committed 58 days ago

Verified e8709167

Older