[Multimodal][Speculative Decoding]Eagle Eagle3 mm support, enablement on qwen2.5vl (#22872)
Signed-off-by: Junhong <liujunhong11@huawei.com> Signed-off-by: Junhong Liu <98734602+LJH-LBJ@users.noreply.github.com> Co-authored-by: Junhong <liujunhong11@huawei.com> Co-authored-by: LJH-LBJ <98734602+LJH-LBJ@users.noreply.github.com>
This commit is contained in:
@@ -651,6 +651,9 @@ _SPECULATIVE_DECODING_EXAMPLE_MODELS = {
|
||||
"MiMoMTPModel": _HfExamplesInfo("XiaomiMiMo/MiMo-7B-RL",
|
||||
trust_remote_code=True,
|
||||
speculative_model="XiaomiMiMo/MiMo-7B-RL"),
|
||||
"Eagle3Qwen2_5vlForCausalLM": _HfExamplesInfo(
|
||||
"Qwen/Qwen2.5-VL-7B-Instruct",
|
||||
speculative_model="Rayzl/qwen2.5-vl-7b-eagle3-sgl"),
|
||||
"Qwen3NextMTP": _HfExamplesInfo("Qwen/Qwen3-Next-80B-A3B-Instruct",
|
||||
min_transformers_version="4.56.3"),
|
||||
}
|
||||
|
||||
@@ -129,6 +129,11 @@ def test_ngram_correctness(
|
||||
["model_setup", "mm_enabled"],
|
||||
[
|
||||
(("eagle3", "Qwen/Qwen3-8B", "AngelSlim/Qwen3-8B_eagle3", 1), False),
|
||||
pytest.param(("eagle3", "Qwen/Qwen2.5-VL-7B-Instruct",
|
||||
"Rayzl/qwen2.5-vl-7b-eagle3-sgl", 1),
|
||||
False,
|
||||
marks=pytest.mark.skip(reason="Skipping due to its " \
|
||||
"head_dim not being a a multiple of 32")),
|
||||
(("eagle", "meta-llama/Llama-3.1-8B-Instruct",
|
||||
"yuhuili/EAGLE-LLaMA3.1-Instruct-8B", 1), False),
|
||||
(("eagle3", "meta-llama/Llama-3.1-8B-Instruct",
|
||||
@@ -145,8 +150,8 @@ def test_ngram_correctness(
|
||||
"eagle618/eagle-deepseek-v3-random", 1), False),
|
||||
],
|
||||
ids=[
|
||||
"qwen3_eagle3", "llama3_eagle", "llama3_eagle3", "llama4_eagle",
|
||||
"llama4_eagle_mm", "deepseek_eagle"
|
||||
"qwen3_eagle3", "qwen2_5_vl_eagle3", "llama3_eagle", "llama3_eagle3",
|
||||
"llama4_eagle", "llama4_eagle_mm", "deepseek_eagle"
|
||||
])
|
||||
@pytest.mark.parametrize("attn_backend",
|
||||
get_attn_backend_list_based_on_platform())
|
||||
|
||||
Reference in New Issue
Block a user