[Quant][Kernel] Remove GPTQ Group/Dynamic Activation Ordering (#54809)

Signed-off-by: Roderick-Wu <roderickwu2003@gmail.com>
Signed-off-by: Roderick Wu <roderickwu2003@gmail.com>
Signed-off-by: mgoin <mgoin64@gmail.com>
Co-authored-by: mgoin <mgoin64@gmail.com>
This commit is contained in:
Roderick Wu
2026-09-08 12:45:12 -07:00
committed by GitHub
co-authored by mgoin
parent 1a522b6949
commit 2c9d68f80b
99 changed files with 1182 additions and 3917 deletions
@@ -22,12 +22,10 @@ os.environ["TOKENIZERS_PARALLELISM"] = "true"
MAX_MODEL_LEN = 1024
MODELS = [
# act_order==True, group_size=128
("TheBloke/TinyLlama-1.1B-Chat-v1.0-GPTQ", "main"),
# 8-bit, act_order==True, group_size=channelwise
("LnL-AI/TinyLlama-1.1B-Chat-v1.0-GPTQ-4bit", "main"),
# desc_act=True is a no-op for channelwise quantization.
("TheBloke/TinyLlama-1.1B-Chat-v1.0-GPTQ", "gptq-8bit--1g-actorder_True"),
# 4-bit, act_order==True, group_size=128
("TechxGenus/gemma-1.1-2b-it-GPTQ", "main"),
("TechxGenus/gemma-2b-it-GPTQ", "main"),
]
+1 -1
View File
@@ -243,7 +243,7 @@ def test_distributed(
"model, quantization_kwargs",
[
("TheBloke/TinyLlama-1.1B-Chat-v0.3-AWQ", {}),
("TheBloke/TinyLlama-1.1B-Chat-v0.3-GPTQ", {}),
("LnL-AI/TinyLlama-1.1B-Chat-v1.0-GPTQ-4bit", {}),
],
)
@pytest.mark.parametrize("max_tokens", [32])