[Model] Pooling models default to using chunked prefill & prefix caching if supported. (#20930)
Signed-off-by: wang.yuqi <noooop@126.com>
This commit is contained in:
@@ -3,42 +3,43 @@
|
||||
|
||||
import pytest
|
||||
|
||||
from .embed_utils import EmbedModelInfo, correctness_test_embed_models
|
||||
from ...utils import CLSPoolingEmbedModelInfo, EmbedModelInfo
|
||||
from .embed_utils import correctness_test_embed_models
|
||||
from .mteb_utils import mteb_test_embed_models
|
||||
|
||||
MODELS = [
|
||||
EmbedModelInfo("Snowflake/snowflake-arctic-embed-xs",
|
||||
is_matryoshka=False,
|
||||
architecture="BertModel",
|
||||
enable_test=True),
|
||||
EmbedModelInfo("Snowflake/snowflake-arctic-embed-s",
|
||||
is_matryoshka=False,
|
||||
architecture="BertModel",
|
||||
enable_test=False),
|
||||
EmbedModelInfo("Snowflake/snowflake-arctic-embed-m",
|
||||
is_matryoshka=False,
|
||||
architecture="BertModel",
|
||||
enable_test=False),
|
||||
EmbedModelInfo("Snowflake/snowflake-arctic-embed-m-long",
|
||||
is_matryoshka=False,
|
||||
architecture="NomicBertModel",
|
||||
enable_test=True),
|
||||
EmbedModelInfo("Snowflake/snowflake-arctic-embed-l",
|
||||
is_matryoshka=False,
|
||||
architecture="BertModel",
|
||||
enable_test=False),
|
||||
EmbedModelInfo("Snowflake/snowflake-arctic-embed-m-v1.5",
|
||||
is_matryoshka=True,
|
||||
architecture="BertModel",
|
||||
enable_test=True),
|
||||
EmbedModelInfo("Snowflake/snowflake-arctic-embed-l-v2.0",
|
||||
is_matryoshka=True,
|
||||
architecture="XLMRobertaModel",
|
||||
enable_test=True),
|
||||
EmbedModelInfo("Snowflake/snowflake-arctic-embed-m-v2.0",
|
||||
is_matryoshka=True,
|
||||
architecture="GteModel",
|
||||
enable_test=True),
|
||||
CLSPoolingEmbedModelInfo("Snowflake/snowflake-arctic-embed-xs",
|
||||
is_matryoshka=False,
|
||||
architecture="BertModel",
|
||||
enable_test=True),
|
||||
CLSPoolingEmbedModelInfo("Snowflake/snowflake-arctic-embed-s",
|
||||
is_matryoshka=False,
|
||||
architecture="BertModel",
|
||||
enable_test=False),
|
||||
CLSPoolingEmbedModelInfo("Snowflake/snowflake-arctic-embed-m",
|
||||
is_matryoshka=False,
|
||||
architecture="BertModel",
|
||||
enable_test=False),
|
||||
CLSPoolingEmbedModelInfo("Snowflake/snowflake-arctic-embed-m-long",
|
||||
is_matryoshka=False,
|
||||
architecture="NomicBertModel",
|
||||
enable_test=True),
|
||||
CLSPoolingEmbedModelInfo("Snowflake/snowflake-arctic-embed-l",
|
||||
is_matryoshka=False,
|
||||
architecture="BertModel",
|
||||
enable_test=False),
|
||||
CLSPoolingEmbedModelInfo("Snowflake/snowflake-arctic-embed-m-v1.5",
|
||||
is_matryoshka=True,
|
||||
architecture="BertModel",
|
||||
enable_test=True),
|
||||
CLSPoolingEmbedModelInfo("Snowflake/snowflake-arctic-embed-l-v2.0",
|
||||
is_matryoshka=True,
|
||||
architecture="XLMRobertaModel",
|
||||
enable_test=True),
|
||||
CLSPoolingEmbedModelInfo("Snowflake/snowflake-arctic-embed-m-v2.0",
|
||||
is_matryoshka=True,
|
||||
architecture="GteModel",
|
||||
enable_test=True),
|
||||
]
|
||||
|
||||
|
||||
|
||||
Reference in New Issue
Block a user