[Quantization] Deprecate Long Tail of Schemes (#31688)

Signed-off-by: Robert Shaw <robshaw@redhat.com>
Signed-off-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com>
Co-authored-by: Robert Shaw <robshaw@redhat.com>
Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
This commit is contained in:
Robert Shaw
2026-01-08 19:07:45 -05:00
committed by GitHub
co-authored by Robert Shaw Wentao Ye
parent d62cfe546d
commit 5825bbc1f7
8 changed files with 61 additions and 5 deletions
+4 -1
View File
@@ -62,7 +62,10 @@ def models_list(*, all: bool = True, keywords: list[str] | None = None):
TEST_MODELS.append(
(
"alexm-nm/tinyllama-24-marlin24-4bit-g128",
{"quantization": "gptq_marlin_24"},
{
"quantization": "gptq_marlin_24",
"allow_deprecated_quantization": True,
},
)
)
@@ -63,7 +63,10 @@ def test_models(
num_logprobs: int,
) -> None:
with vllm_runner(
model_pair.model_marlin, dtype=dtype, quantization="gptq_marlin_24"
model_pair.model_marlin,
dtype=dtype,
quantization="gptq_marlin_24",
allow_deprecated_quantization=True,
) as marlin_24_model:
marlin_24_outputs = marlin_24_model.generate_greedy_logprobs(
example_prompts, max_tokens, num_logprobs
+3 -1
View File
@@ -26,7 +26,9 @@ MODELS = [
)
@pytest.mark.parametrize("model", MODELS)
def test_auto_round(vllm_runner, model):
with vllm_runner(model, enforce_eager=True) as llm:
with vllm_runner(
model, enforce_eager=True, allow_deprecated_quantization=True
) as llm:
output = llm.generate_greedy(["The capital of France is"], max_tokens=8)
assert output
print(f"{output[0][1]}")
+5 -1
View File
@@ -34,6 +34,10 @@ def test_model_experts_int8_startup(
model_info.check_transformers_version(on_fail="skip")
with vllm_runner(
model, dtype=dtype, enforce_eager=True, quantization="experts_int8"
model,
dtype=dtype,
enforce_eager=True,
quantization="experts_int8",
allow_deprecated_quantization=True,
) as vllm_model:
vllm_model.generate_greedy(example_prompts, max_tokens)
+5 -1
View File
@@ -30,6 +30,10 @@ def test_model_rtn_startup(
max_tokens: int,
) -> None:
with vllm_runner(
model, enforce_eager=True, dtype=dtype, quantization="rtn"
model,
enforce_eager=True,
dtype=dtype,
quantization="rtn",
allow_deprecated_quantization=True,
) as vllm_model:
vllm_model.generate_greedy(example_prompts, max_tokens)