biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Renderer] Separate out RendererConfig from ModelConfig (#30145)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-12-07 15:15:42 +08:00
committed by GitHub
parent a49d813fa8
commit 27f4c2fd46
105 changed files with 969 additions and 797 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
tests/model_executor/test_model_load_with_params.py
View File

@@ -42,8 +42,10 @@ def test_model_loading_with_params(vllm_runner, monkeypatch):
"Write a short story about a robot that dreams for the first time.\n"
)
model_config = vllm_model.llm.llm_engine.model_config
model_tokenizer = vllm_model.llm.llm_engine.tokenizer
llm_engine = vllm_model.llm.llm_engine
model_config = llm_engine.model_config
renderer_config = llm_engine.renderer_config
tokenizer = llm_engine.tokenizer
# asserts on the bert model config file
assert model_config.encoder_config["max_seq_length"] == 512
@@ -54,8 +56,8 @@ def test_model_loading_with_params(vllm_runner, monkeypatch):
assert model_config.pooler_config.normalize
# asserts on the tokenizer loaded
assert model_config.tokenizer == "BAAI/bge-base-en-v1.5"
assert model_tokenizer.model_max_length == 512
assert renderer_config.tokenizer == "BAAI/bge-base-en-v1.5"
assert tokenizer.model_max_length == 512
def check_model(model):
assert isinstance(model, BertEmbeddingModel)
@@ -86,8 +88,10 @@ def test_roberta_model_loading_with_params(vllm_runner, monkeypatch):
"Write a short story about a robot that dreams for the first time.\n"
)
model_config = vllm_model.llm.llm_engine.model_config
model_tokenizer = vllm_model.llm.llm_engine.tokenizer
llm_engine = vllm_model.llm.llm_engine
model_config = llm_engine.model_config
renderer_config = llm_engine.renderer_config
tokenizer = llm_engine.tokenizer
# asserts on the bert model config file
assert model_config.encoder_config["max_seq_length"] == 512
@@ -98,8 +102,8 @@ def test_roberta_model_loading_with_params(vllm_runner, monkeypatch):
assert model_config.pooler_config.normalize
# asserts on the tokenizer loaded
assert model_config.tokenizer == "intfloat/multilingual-e5-base"
assert model_tokenizer.model_max_length == 512
assert renderer_config.tokenizer == "intfloat/multilingual-e5-base"
assert tokenizer.model_max_length == 512
def check_model(model):
assert isinstance(model, RobertaEmbeddingModel)
@@ -128,7 +132,7 @@ def test_facebook_roberta_model_loading_with_params(vllm_runner, monkeypatch):
"Write a short story about a robot that dreams for the first time.\n"
)
assert vllm_model.llm.llm_engine.model_config.tokenizer == model_name
assert vllm_model.llm.llm_engine.renderer_config.tokenizer == model_name
def check_model(model):
assert isinstance(model, RobertaEmbeddingModel)