xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 00:25:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[BugFix] [Spec Decode] Remove LlamaForCausalLMEagle3 to fix CI (#22611)

Browse Source 
Signed-off-by: 22quinn <33176974+22quinn@users.noreply.github.com>
This commit is contained in:
22quinn 2025-08-11 10:31:36 -07:00 committed by GitHub
parent c90fb03df5
commit 807d21b80d
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
4 changed files with 32 additions and 27 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
tests/models/registry.py
View File

@ -525,10 +525,11 @@ _SPECULATIVE_DECODING_EXAMPLE_MODELS = {
trust_remote_code=True, trust_remote_code=True,
speculative_model="yuhuili/EAGLE3-LLaMA3.1-Instruct-8B", speculative_model="yuhuili/EAGLE3-LLaMA3.1-Instruct-8B",
tokenizer="meta-llama/Llama-3.1-8B-Instruct"), tokenizer="meta-llama/Llama-3.1-8B-Instruct"),
"LlamaForCausalLMEagle3": _HfExamplesInfo("AngelSlim/Qwen3-8B_eagle3", # noqa: E501 # TODO: Re-enable this once tests/models/test_initialization.py is fixed, see PR #22333 #22611 # noqa: E501
trust_remote_code=True, # "LlamaForCausalLMEagle3": _HfExamplesInfo("AngelSlim/Qwen3-8B_eagle3", # noqa: E501
speculative_model="AngelSlim/Qwen3-8B_eagle3", # trust_remote_code=True,
tokenizer="Qwen/Qwen3-8B"), # speculative_model="AngelSlim/Qwen3-8B_eagle3", # noqa: E501
# tokenizer="Qwen/Qwen3-8B"),
"EagleLlama4ForCausalLM": _HfExamplesInfo( "EagleLlama4ForCausalLM": _HfExamplesInfo(
"morgendave/EAGLE-Llama-4-Scout-17B-16E-Instruct", "morgendave/EAGLE-Llama-4-Scout-17B-16E-Instruct",
trust_remote_code=True, trust_remote_code=True,

45
tests/v1/e2e/test_spec_decode.py
View File

@ -125,27 +125,30 @@ def test_ngram_correctness(
cleanup_dist_env_and_memory() cleanup_dist_env_and_memory()
@pytest.mark.parametrize(["model_setup", "mm_enabled"], [ @pytest.mark.parametrize(
(("eagle3", "Qwen/Qwen3-8B", "AngelSlim/Qwen3-8B_eagle3", 1), False), ["model_setup", "mm_enabled"],
(("eagle", "meta-llama/Llama-3.1-8B-Instruct", [
"yuhuili/EAGLE-LLaMA3.1-Instruct-8B", 1), False), # TODO: Re-enable this once tests/models/test_initialization.py is fixed, see PR #22333 #22611 # noqa: E501
(("eagle3", "meta-llama/Llama-3.1-8B-Instruct", # (("eagle3", "Qwen/Qwen3-8B", "AngelSlim/Qwen3-8B_eagle3", 1), False),
"yuhuili/EAGLE3-LLaMA3.1-Instruct-8B", 1), False), (("eagle", "meta-llama/Llama-3.1-8B-Instruct",
pytest.param( "yuhuili/EAGLE-LLaMA3.1-Instruct-8B", 1), False),
("eagle", "meta-llama/Llama-4-Scout-17B-16E-Instruct", (("eagle3", "meta-llama/Llama-3.1-8B-Instruct",
"morgendave/EAGLE-Llama-4-Scout-17B-16E-Instruct", 4), "yuhuili/EAGLE3-LLaMA3.1-Instruct-8B", 1), False),
False, pytest.param(
marks=pytest.mark.skip(reason="Skipping due to CI OOM issues")), ("eagle", "meta-llama/Llama-4-Scout-17B-16E-Instruct",
pytest.param( "morgendave/EAGLE-Llama-4-Scout-17B-16E-Instruct", 4),
("eagle", "meta-llama/Llama-4-Scout-17B-16E-Instruct", False,
"morgendave/EAGLE-Llama-4-Scout-17B-16E-Instruct", 4), marks=pytest.mark.skip(reason="Skipping due to CI OOM issues")),
True, pytest.param(
marks=pytest.mark.skip(reason="Skipping due to CI OOM issues")), ("eagle", "meta-llama/Llama-4-Scout-17B-16E-Instruct",
], "morgendave/EAGLE-Llama-4-Scout-17B-16E-Instruct", 4),
ids=[ True,
"qwen3_eagle3", "llama3_eagle", "llama3_eagle3", marks=pytest.mark.skip(reason="Skipping due to CI OOM issues")),
"llama4_eagle", "llama4_eagle_mm" ],
]) ids=[
"qwen3_eagle3", "llama3_eagle", "llama3_eagle3", "llama4_eagle",
"llama4_eagle_mm"
])
@pytest.mark.parametrize("attn_backend", @pytest.mark.parametrize("attn_backend",
get_attn_backend_list_based_on_platform()) get_attn_backend_list_based_on_platform())
def test_eagle_correctness( def test_eagle_correctness(

3
vllm/model_executor/models/registry.py
View File

@ -259,7 +259,8 @@ _SPECULATIVE_DECODING_MODELS = {
"EagleLlama4ForCausalLM": ("llama4_eagle", "EagleLlama4ForCausalLM"), "EagleLlama4ForCausalLM": ("llama4_eagle", "EagleLlama4ForCausalLM"),
"EagleMiniCPMForCausalLM": ("minicpm_eagle", "EagleMiniCPMForCausalLM"), "EagleMiniCPMForCausalLM": ("minicpm_eagle", "EagleMiniCPMForCausalLM"),
"Eagle3LlamaForCausalLM": ("llama_eagle3", "Eagle3LlamaForCausalLM"), "Eagle3LlamaForCausalLM": ("llama_eagle3", "Eagle3LlamaForCausalLM"),
"LlamaForCausalLMEagle3": ("llama_eagle3", "Eagle3LlamaForCausalLM"), # TODO: Re-enable this once tests/models/test_initialization.py is fixed, see PR #22333 #22611 # noqa: E501
# "LlamaForCausalLMEagle3": ("llama_eagle3", "Eagle3LlamaForCausalLM"),
"DeepSeekMTPModel": ("deepseek_mtp", "DeepSeekMTP"), "DeepSeekMTPModel": ("deepseek_mtp", "DeepSeekMTP"),
"Glm4MoeMTPModel": ("glm4_moe_mtp", "Glm4MoeMTP"), "Glm4MoeMTPModel": ("glm4_moe_mtp", "Glm4MoeMTP"),
"MedusaModel": ("medusa", "Medusa"), "MedusaModel": ("medusa", "Medusa"),

2
vllm/transformers_utils/configs/eagle.py
View File

@ -45,7 +45,7 @@ class EAGLEConfig(PretrainedConfig):
# Eagle model name should follow naming convention of # Eagle model name should follow naming convention of
# LlamaForCausalLM -> EagleLlamaForCausalLM # LlamaForCausalLM -> EagleLlamaForCausalLM
# LlamaForCausalLM -> Eagle3LlamaForCausalLM / LlamaForCausalLMEagle3 # LlamaForCausalLM -> Eagle3LlamaForCausalLM
if method == "eagle": if method == "eagle":
assert self.model is not None, \ assert self.model is not None, \
"model should not be None when method is eagle" "model should not be None when method is eagle"