Skip to content

add ut for gguf alg_ext and update so file #1012

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Nov 11, 2025
Merged

add ut for gguf alg_ext and update so file #1012

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Binary file modified auto_round/alg_ext.abi3.so
View file
Open in desktop
Binary file not shown.
6 changes: 4 additions & 2 deletions auto_round/compressors/base.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -698,11 +698,13 @@ def _check_compatibility(self) -> None:
has_besides_gguf = True
if has_gguf and has_besides_gguf:
raise ValueError("Gguf format is not compatible with other formats, please choose only one of them")
if has_gguf and self.iters != 0 and self.bits != 3:
if has_gguf and self.iters != 0 and self.bits != 3 and not self.enable_alg_ext:
logger.warning(
"`iters=0` is recommended when exporting to GGUF format except for bits 3,"
" as we have optimized the RTN method for this case."
" We are likely to release new algorithm for certain configurations in the future."
" Or add enable_alg_ext to use the new algorithm,"
" refer to https://github.com/intel/auto-round/tree/main/docs/gguf_alg_ext_acc.md"
" to check the acc."
)

if (
Expand Down
16 changes: 16 additions & 0 deletions docs/gguf_alg_ext_acc.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,16 @@
We use **lm-eval** for evaluation. For LLaMA, we enabled `add_bos_token` and
`removed @use_kernel_forward_from_hub("RMSNorm")`
in [modeling_llama.py](https://github.com/huggingface/transformers/blob/main/src/transformers/models/llama/modeling_llama.py#L52C1-L52C40)
to stabilize accuracy during evaluation. All other settings follow the default configurations of AutoRound and lm-eval.

*Average accuracy across `lambada_openai`, `hellaswag`, `piqa`, `winogrande`, `truthfulqa_mc1`, `openbookqa`, `boolq`, `arc_easy`, `arc_challenge` and `mmlu`.*

|method|scheme|Llama-3.1-8B|Qwen2.5-7B-Instruct|Qwen3-8b|Qwen3-30B-A3B-Instruct-2507|
|:-----|:-----|:-----------|:------------------|:-------|:--------------------------|
|**BF16** | - |0.6295(100%)|0.6571(100%) |0.6322(100%)|0.6746(100%) |
| **original** | q2_k_s | 0.5535(87.92%)| 0.6266(95.35%)|0.5901(93.35%)|0.6386(94.66%)|
| **enable_alg_ext** |q2_k_s|0.5740(91.18%)|0.6349(96.62%)|0.5962(94.31%)|0.6460(95.77%)|
| **original** | q3_k_s | 0.6040(95.95%)|0.6382(97.12%)|0.6128(96.94%)|0.6598(97.82%)|
| **enable_alg_ext** |q3_k_s|0.6081(96.59%)|0.6503(98.97%)|0.6252(98.89%)|0.6622(98.17%)|
| **original** | q4_k_s | 0.6228(98.94%)|0.6560(99.83%)|0.6303(99.70%)|0.6762(100.24%)|
| **enable_alg_ext** |q4_k_s|0.6239(99.11%)|0.6605(100.51%)|0.6320(99.98%)|0.6777(100.46%)|
6 changes: 5 additions & 1 deletion test/test_cpu/test_autoround.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -716,8 +716,12 @@ def test_alg_ext(self):
ar = AutoRound(model_name, scheme="W2A16", iters=1, nsamples=1, enable_alg_ext=True)
ar.quantize()

model_name = "/tf_dataset/auto_round/models/Qwen/Qwen3-0.6B"
ar = AutoRound(model_name, scheme="gguf:q4_k_s", iters=1, nsamples=1, enable_alg_ext=True)
ar.quantize()

def test_alg_ext_import(self):
from auto_round.alg_ext import quantize_block_ext
from auto_round.alg_ext import dq_quantize_block_ext, quantize_block_ext

def test_invalid_layer_config(self):
with self.assertRaises(ValueError):
Expand Down