enhance LoRA tests and fix base model LoRA eviction in Scheduler (#16333)
This commit is contained in:
@@ -17,29 +17,11 @@ import unittest
|
||||
|
||||
from sglang.test.lora_utils import (
|
||||
CI_MULTI_LORA_MODELS,
|
||||
LoRAAdaptor,
|
||||
LoRAModelCase,
|
||||
LORA_MODELS_QWEN3,
|
||||
run_lora_multiple_batch_on_model_cases,
|
||||
)
|
||||
from sglang.test.test_utils import CustomTestCase
|
||||
|
||||
LORA_MODELS_QWEN3 = [
|
||||
LoRAModelCase(
|
||||
base="Qwen/Qwen3-4B",
|
||||
adaptors=[
|
||||
LoRAAdaptor(
|
||||
name="nissenj/Qwen3-4B-lora-v2",
|
||||
prefill_tolerance=3e-1,
|
||||
),
|
||||
LoRAAdaptor(
|
||||
name="y9760210/Qwen3-4B-lora_model",
|
||||
prefill_tolerance=3e-1,
|
||||
),
|
||||
],
|
||||
max_loras_per_batch=2,
|
||||
),
|
||||
]
|
||||
|
||||
|
||||
class TestLoRASpecDecoding(CustomTestCase):
|
||||
def test_qwen(self):
|
||||
|
||||
Reference in New Issue
Block a user