diff --git a/tests/models/registry.py b/tests/models/registry.py index 898e38a4ae1d..c5816df25b96 100644 --- a/tests/models/registry.py +++ b/tests/models/registry.py @@ -525,10 +525,11 @@ def check_available_online( trust_remote_code=True, speculative_model="yuhuili/EAGLE3-LLaMA3.1-Instruct-8B", tokenizer="meta-llama/Llama-3.1-8B-Instruct"), - "LlamaForCausalLMEagle3": _HfExamplesInfo("AngelSlim/Qwen3-8B_eagle3", # noqa: E501 - trust_remote_code=True, - speculative_model="AngelSlim/Qwen3-8B_eagle3", - tokenizer="Qwen/Qwen3-8B"), + # TODO: Re-enable this once tests/models/test_initialization.py is fixed, see PR #22333 #22611 # noqa: E501 + # "LlamaForCausalLMEagle3": _HfExamplesInfo("AngelSlim/Qwen3-8B_eagle3", # noqa: E501 + # trust_remote_code=True, + # speculative_model="AngelSlim/Qwen3-8B_eagle3", # noqa: E501 + # tokenizer="Qwen/Qwen3-8B"), "EagleLlama4ForCausalLM": _HfExamplesInfo( "morgendave/EAGLE-Llama-4-Scout-17B-16E-Instruct", trust_remote_code=True, diff --git a/tests/v1/e2e/test_spec_decode.py b/tests/v1/e2e/test_spec_decode.py index cd383b58db2a..599916c0d1cf 100644 --- a/tests/v1/e2e/test_spec_decode.py +++ b/tests/v1/e2e/test_spec_decode.py @@ -125,27 +125,30 @@ def test_ngram_correctness( cleanup_dist_env_and_memory() -@pytest.mark.parametrize(["model_setup", "mm_enabled"], [ - (("eagle3", "Qwen/Qwen3-8B", "AngelSlim/Qwen3-8B_eagle3", 1), False), - (("eagle", "meta-llama/Llama-3.1-8B-Instruct", - "yuhuili/EAGLE-LLaMA3.1-Instruct-8B", 1), False), - (("eagle3", "meta-llama/Llama-3.1-8B-Instruct", - "yuhuili/EAGLE3-LLaMA3.1-Instruct-8B", 1), False), - pytest.param( - ("eagle", "meta-llama/Llama-4-Scout-17B-16E-Instruct", - "morgendave/EAGLE-Llama-4-Scout-17B-16E-Instruct", 4), - False, - marks=pytest.mark.skip(reason="Skipping due to CI OOM issues")), - pytest.param( - ("eagle", "meta-llama/Llama-4-Scout-17B-16E-Instruct", - "morgendave/EAGLE-Llama-4-Scout-17B-16E-Instruct", 4), - True, - marks=pytest.mark.skip(reason="Skipping due to CI OOM issues")), -], - ids=[ - "qwen3_eagle3", "llama3_eagle", "llama3_eagle3", - "llama4_eagle", "llama4_eagle_mm" - ]) +@pytest.mark.parametrize( + ["model_setup", "mm_enabled"], + [ + # TODO: Re-enable this once tests/models/test_initialization.py is fixed, see PR #22333 #22611 # noqa: E501 + # (("eagle3", "Qwen/Qwen3-8B", "AngelSlim/Qwen3-8B_eagle3", 1), False), + (("eagle", "meta-llama/Llama-3.1-8B-Instruct", + "yuhuili/EAGLE-LLaMA3.1-Instruct-8B", 1), False), + (("eagle3", "meta-llama/Llama-3.1-8B-Instruct", + "yuhuili/EAGLE3-LLaMA3.1-Instruct-8B", 1), False), + pytest.param( + ("eagle", "meta-llama/Llama-4-Scout-17B-16E-Instruct", + "morgendave/EAGLE-Llama-4-Scout-17B-16E-Instruct", 4), + False, + marks=pytest.mark.skip(reason="Skipping due to CI OOM issues")), + pytest.param( + ("eagle", "meta-llama/Llama-4-Scout-17B-16E-Instruct", + "morgendave/EAGLE-Llama-4-Scout-17B-16E-Instruct", 4), + True, + marks=pytest.mark.skip(reason="Skipping due to CI OOM issues")), + ], + ids=[ + "qwen3_eagle3", "llama3_eagle", "llama3_eagle3", "llama4_eagle", + "llama4_eagle_mm" + ]) @pytest.mark.parametrize("attn_backend", get_attn_backend_list_based_on_platform()) def test_eagle_correctness( diff --git a/vllm/model_executor/models/registry.py b/vllm/model_executor/models/registry.py index aca3d84f0071..1d387e64be72 100644 --- a/vllm/model_executor/models/registry.py +++ b/vllm/model_executor/models/registry.py @@ -259,7 +259,8 @@ "EagleLlama4ForCausalLM": ("llama4_eagle", "EagleLlama4ForCausalLM"), "EagleMiniCPMForCausalLM": ("minicpm_eagle", "EagleMiniCPMForCausalLM"), "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"), "Glm4MoeMTPModel": ("glm4_moe_mtp", "Glm4MoeMTP"), "MedusaModel": ("medusa", "Medusa"), diff --git a/vllm/transformers_utils/configs/eagle.py b/vllm/transformers_utils/configs/eagle.py index 01217eb19126..bc249c583603 100644 --- a/vllm/transformers_utils/configs/eagle.py +++ b/vllm/transformers_utils/configs/eagle.py @@ -45,7 +45,7 @@ def __init__(self, # Eagle model name should follow naming convention of # LlamaForCausalLM -> EagleLlamaForCausalLM - # LlamaForCausalLM -> Eagle3LlamaForCausalLM / LlamaForCausalLMEagle3 + # LlamaForCausalLM -> Eagle3LlamaForCausalLM if method == "eagle": assert self.model is not None, \ "model should not be None when method is eagle"