[CORE] Prompt Embeddings Support for v1 Engine (#24278)

Signed-off-by: Andrew Sansom <andrew@protopia.ai>
Signed-off-by: Andrew Sansom <qthequartermasterman@gmail.com>
Co-authored-by: Cyrus Leung <cyrus.tl.leung@gmail.com>
This commit is contained in:
Andrew Sansom
2025-09-19 08:03:09 +08:00
committed by GitHub
co-authored by Cyrus Leung
parent 9fac6aa30b
commit 9a4600e4dc
20 changed files with 305 additions and 76 deletions
@@ -76,11 +76,6 @@ def test_models(
model_executor: str,
enable_prompt_embeds: bool,
) -> None:
if enable_prompt_embeds and envs.is_set(
"VLLM_USE_V1") and envs.VLLM_USE_V1:
pytest.skip("enable_prompt_embeds is not supported in v1.")
if not envs.VLLM_USE_V1:
if async_scheduling:
pytest.skip("async_scheduling only supported in v1.")
@@ -164,11 +159,6 @@ def test_models_distributed(
extra_env: dict[str, str],
enable_prompt_embeds: bool,
) -> None:
if enable_prompt_embeds and envs.is_set(
"VLLM_USE_V1") and envs.VLLM_USE_V1:
pytest.skip("enable_prompt_embeds is not supported in v1.")
if test_suite != TARGET_TEST_SUITE:
pytest.skip(f"Skip test for {test_suite}")
@@ -36,7 +36,6 @@ def default_server_args() -> list[str]:
"--enforce-eager",
# Prompt Embeds server args
"--enable-prompt-embeds",
"--no-enable-chunked-prefill",
]
@@ -125,12 +125,6 @@ def test_models(hf_runner, vllm_runner, example_prompts, model: str,
# in parts of the operators
pytest.skip(f"Skipping '{model}' model test with AITER kernel.")
# Note: can be removed when
# https://github.com/vllm-project/vllm/pull/24278 finished
if current_platform.is_cpu() and use_prompt_embeds:
pytest.skip("Skipping use_prompt_embeds=True with "
"V1-only CPU backend.")
with hf_runner(model) as hf_model:
hf_outputs = hf_model.generate_greedy_logprobs_limit(
example_prompts, max_tokens, num_logprobs)