[CI] Migrate LoRA tests to test/registered/lora/ (#15176)
This commit is contained in:
Alison Shao
@@ -1,193 +0,0 @@
|
||||
# Copyright 2023-2024 SGLang Team
|
||||
# Licensed under the Apache License, Version 2.0 (the "License");
|
||||
# you may not use this file except in compliance with the License.
|
||||
# You may obtain a copy of the License at
|
||||
#
|
||||
# http://www.apache.org/licenses/LICENSE-2.0
|
||||
#
|
||||
# Unless required by applicable law or agreed to in writing, software
|
||||
# distributed under the License is distributed on an "AS IS" BASIS,
|
||||
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||
# See the License for the specific language governing permissions and
|
||||
# limitations under the License.
|
||||
# ==============================================================================
|
||||
|
||||
"""
|
||||
Unit tests for LoRA eviction policies.
|
||||
Tests LRU and FIFO eviction behavior.
|
||||
"""
|
||||
|
||||
import unittest
|
||||
|
||||
from sglang.srt.lora.eviction_policy import get_eviction_policy
|
||||
from sglang.test.ci.ci_register import register_cuda_ci
|
||||
|
||||
register_cuda_ci(est_time=200, suite="nightly-1-gpu", nightly=True)
|
||||
|
||||
|
||||
class TestLoRAEvictionPolicy(unittest.TestCase):
|
||||
"""Unit tests for LoRA eviction policies."""
|
||||
|
||||
def _test_eviction_policy(
|
||||
self, policy_name, access_sequence, candidates, expected_victim
|
||||
):
|
||||
"""
|
||||
Helper to test eviction policy with given access pattern.
|
||||
|
||||
Args:
|
||||
policy_name: Name of eviction policy ("lru" or "fifo")
|
||||
access_sequence: List of adapter IDs in access order
|
||||
candidates: Set of adapter IDs that can be evicted
|
||||
expected_victim: Expected adapter ID to be evicted
|
||||
"""
|
||||
policy = get_eviction_policy(policy_name)
|
||||
|
||||
# Simulate access pattern
|
||||
for adapter_id in access_sequence:
|
||||
policy.mark_used(adapter_id)
|
||||
|
||||
# Select victim from candidates
|
||||
victim = policy.select_victim(candidates)
|
||||
self.assertEqual(
|
||||
victim,
|
||||
expected_victim,
|
||||
f"{policy_name.upper()}: Expected {expected_victim}, got {victim}",
|
||||
)
|
||||
|
||||
def test_lru_basic(self):
|
||||
"""Test LRU selects least recently used adapter."""
|
||||
self._test_eviction_policy(
|
||||
"lru",
|
||||
access_sequence=["lora1", "lora2", "lora3", "lora4"],
|
||||
candidates={"lora1", "lora2", "lora3", "lora4"},
|
||||
expected_victim="lora1",
|
||||
)
|
||||
|
||||
def test_lru_with_reuse(self):
|
||||
"""Test LRU updates order on reuse."""
|
||||
self._test_eviction_policy(
|
||||
"lru",
|
||||
access_sequence=["lora1", "lora2", "lora3", "lora4", "lora1"],
|
||||
candidates={"lora1", "lora2", "lora3", "lora4"},
|
||||
expected_victim="lora2",
|
||||
)
|
||||
|
||||
def test_lru_multiple_reuse(self):
|
||||
"""Test LRU with multiple reuses."""
|
||||
self._test_eviction_policy(
|
||||
"lru",
|
||||
access_sequence=["lora1", "lora2", "lora3", "lora1", "lora2"],
|
||||
candidates={"lora1", "lora2", "lora3"},
|
||||
expected_victim="lora3",
|
||||
)
|
||||
|
||||
def test_lru_with_subset_candidates(self):
|
||||
"""Test LRU with subset of candidates."""
|
||||
self._test_eviction_policy(
|
||||
"lru",
|
||||
access_sequence=["lora1", "lora2", "lora3", "lora4"],
|
||||
candidates={"lora2", "lora3", "lora4"},
|
||||
expected_victim="lora2",
|
||||
)
|
||||
|
||||
def test_lru_base_model_evicted_last(self):
|
||||
"""Test LRU evicts LoRA adapters before base model (None)."""
|
||||
self._test_eviction_policy(
|
||||
"lru",
|
||||
access_sequence=["lora1", "lora2", "lora3"],
|
||||
candidates={None, "lora1", "lora2", "lora3"},
|
||||
expected_victim="lora1",
|
||||
)
|
||||
|
||||
def test_fifo_basic(self):
|
||||
"""Test FIFO selects first inserted adapter."""
|
||||
self._test_eviction_policy(
|
||||
"fifo",
|
||||
access_sequence=["lora1", "lora2", "lora3", "lora4"],
|
||||
candidates={"lora1", "lora2", "lora3", "lora4"},
|
||||
expected_victim="lora1",
|
||||
)
|
||||
|
||||
def test_fifo_ignores_reuse(self):
|
||||
"""Test FIFO ignores reuse."""
|
||||
self._test_eviction_policy(
|
||||
"fifo",
|
||||
access_sequence=[
|
||||
"lora1",
|
||||
"lora2",
|
||||
"lora3",
|
||||
"lora4",
|
||||
"lora4",
|
||||
"lora3",
|
||||
"lora2",
|
||||
"lora1",
|
||||
],
|
||||
candidates={"lora1", "lora2", "lora3", "lora4"},
|
||||
expected_victim="lora1",
|
||||
)
|
||||
|
||||
def test_fifo_with_subset_candidates(self):
|
||||
"""Test FIFO with subset of candidates."""
|
||||
self._test_eviction_policy(
|
||||
"fifo",
|
||||
access_sequence=["lora1", "lora2", "lora3", "lora4"],
|
||||
candidates={"lora2", "lora3", "lora4"},
|
||||
expected_victim="lora2",
|
||||
)
|
||||
|
||||
def test_fifo_base_model_evicted_last(self):
|
||||
"""Test FIFO evicts LoRA adapters before base model (None)."""
|
||||
self._test_eviction_policy(
|
||||
"fifo",
|
||||
access_sequence=["lora1", "lora2", "lora3"],
|
||||
candidates={None, "lora1", "lora2", "lora3"},
|
||||
expected_victim="lora1",
|
||||
)
|
||||
|
||||
def test_policy_remove(self):
|
||||
"""Test that remove() correctly updates internal state."""
|
||||
lru = get_eviction_policy("lru")
|
||||
lru.mark_used("lora1")
|
||||
lru.mark_used("lora2")
|
||||
lru.mark_used("lora3")
|
||||
|
||||
# Remove lora1, so lora2 becomes LRU
|
||||
lru.remove("lora1")
|
||||
victim = lru.select_victim({"lora1", "lora2", "lora3"})
|
||||
self.assertEqual(victim, "lora2")
|
||||
|
||||
def test_eviction_policy_factory(self):
|
||||
"""Test eviction policy factory function."""
|
||||
# Test valid policies
|
||||
lru = get_eviction_policy("lru")
|
||||
fifo = get_eviction_policy("fifo")
|
||||
|
||||
self.assertIsNotNone(lru)
|
||||
self.assertIsNotNone(fifo)
|
||||
|
||||
# Test invalid policy
|
||||
with self.assertRaises(ValueError):
|
||||
get_eviction_policy("invalid_policy")
|
||||
|
||||
def test_lru_vs_fifo_behavior(self):
|
||||
"""Test that LRU and FIFO behave differently."""
|
||||
access_sequence = ["lora1", "lora2", "lora3", "lora1"]
|
||||
candidates = {"lora1", "lora2", "lora3"}
|
||||
|
||||
lru = get_eviction_policy("lru")
|
||||
for adapter_id in access_sequence:
|
||||
lru.mark_used(adapter_id)
|
||||
lru_victim = lru.select_victim(candidates)
|
||||
|
||||
fifo = get_eviction_policy("fifo")
|
||||
for adapter_id in access_sequence:
|
||||
fifo.mark_used(adapter_id)
|
||||
fifo_victim = fifo.select_victim(candidates)
|
||||
|
||||
self.assertNotEqual(lru_victim, fifo_victim)
|
||||
self.assertEqual(lru_victim, "lora2")
|
||||
self.assertEqual(fifo_victim, "lora1")
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main(verbosity=2)
|
||||
@@ -1,330 +0,0 @@
|
||||
"""
|
||||
Unit tests for OpenAI-compatible LoRA API support.
|
||||
|
||||
Tests the model parameter parsing and LoRA adapter resolution logic
|
||||
that enables OpenAI-compatible LoRA adapter selection.
|
||||
"""
|
||||
|
||||
import unittest
|
||||
from unittest.mock import MagicMock
|
||||
|
||||
from sglang.srt.entrypoints.openai.serving_base import OpenAIServingBase
|
||||
from sglang.test.ci.ci_register import register_cuda_ci
|
||||
|
||||
register_cuda_ci(est_time=30, suite="nightly-1-gpu", nightly=True)
|
||||
from sglang.srt.server_args import ServerArgs
|
||||
|
||||
|
||||
class MockTokenizerManager:
|
||||
"""Mock TokenizerManager for testing."""
|
||||
|
||||
def __init__(self, enable_lora=False):
|
||||
self.server_args = MagicMock(spec=ServerArgs)
|
||||
self.server_args.enable_lora = enable_lora
|
||||
self.server_args.tokenizer_metrics_allowed_custom_labels = None
|
||||
|
||||
|
||||
class ConcreteServingBase(OpenAIServingBase):
|
||||
"""Concrete implementation for testing abstract base class."""
|
||||
|
||||
def _request_id_prefix(self) -> str:
|
||||
return "test-"
|
||||
|
||||
def _convert_to_internal_request(self, request, raw_request=None):
|
||||
pass
|
||||
|
||||
def _validate_request(self, request):
|
||||
pass
|
||||
|
||||
|
||||
class TestParseModelParameter(unittest.TestCase):
|
||||
"""Test _parse_model_parameter method."""
|
||||
|
||||
def setUp(self):
|
||||
self.tokenizer_manager = MockTokenizerManager(enable_lora=True)
|
||||
self.serving = ConcreteServingBase(self.tokenizer_manager)
|
||||
|
||||
def test_model_without_adapter(self):
|
||||
"""Test parsing model without adapter returns None for adapter."""
|
||||
base_model, adapter = self.serving._parse_model_parameter("llama-3.1-8B")
|
||||
self.assertEqual(base_model, "llama-3.1-8B")
|
||||
self.assertIsNone(adapter)
|
||||
|
||||
def test_model_with_adapter(self):
|
||||
"""Test parsing model with adapter extracts both parts."""
|
||||
base_model, adapter = self.serving._parse_model_parameter(
|
||||
"llama-3.1-8B:sql-expert"
|
||||
)
|
||||
self.assertEqual(base_model, "llama-3.1-8B")
|
||||
self.assertEqual(adapter, "sql-expert")
|
||||
|
||||
def test_model_with_path_and_adapter(self):
|
||||
"""Test parsing model path with slashes and adapter."""
|
||||
base_model, adapter = self.serving._parse_model_parameter(
|
||||
"meta-llama/Llama-3.1-8B-Instruct:adapter-name"
|
||||
)
|
||||
self.assertEqual(base_model, "meta-llama/Llama-3.1-8B-Instruct")
|
||||
self.assertEqual(adapter, "adapter-name")
|
||||
|
||||
def test_model_with_multiple_colons(self):
|
||||
"""Test that only first colon is used for splitting."""
|
||||
base_model, adapter = self.serving._parse_model_parameter("model:adapter:extra")
|
||||
self.assertEqual(base_model, "model")
|
||||
self.assertEqual(adapter, "adapter:extra")
|
||||
|
||||
def test_model_with_whitespace(self):
|
||||
"""Test that whitespace is stripped from both parts."""
|
||||
base_model, adapter = self.serving._parse_model_parameter(
|
||||
" model-name : adapter-name "
|
||||
)
|
||||
self.assertEqual(base_model, "model-name")
|
||||
self.assertEqual(adapter, "adapter-name")
|
||||
|
||||
def test_model_with_empty_adapter(self):
|
||||
"""Test model ending with colon returns None for adapter."""
|
||||
base_model, adapter = self.serving._parse_model_parameter("model-name:")
|
||||
self.assertEqual(base_model, "model-name")
|
||||
self.assertIsNone(adapter)
|
||||
|
||||
def test_model_with_only_spaces_after_colon(self):
|
||||
"""Test model with only whitespace after colon returns None for adapter."""
|
||||
base_model, adapter = self.serving._parse_model_parameter("model-name: ")
|
||||
self.assertEqual(base_model, "model-name")
|
||||
self.assertIsNone(adapter)
|
||||
|
||||
|
||||
class TestResolveLoraPath(unittest.TestCase):
|
||||
"""Test _resolve_lora_path method."""
|
||||
|
||||
def setUp(self):
|
||||
self.tokenizer_manager = MockTokenizerManager(enable_lora=True)
|
||||
self.serving = ConcreteServingBase(self.tokenizer_manager)
|
||||
|
||||
def test_no_adapter_specified(self):
|
||||
"""Test when neither model nor explicit lora_path has adapter."""
|
||||
result = self.serving._resolve_lora_path("model-name", None)
|
||||
self.assertIsNone(result)
|
||||
|
||||
def test_adapter_in_model_only(self):
|
||||
"""Test adapter from model parameter when no explicit path."""
|
||||
result = self.serving._resolve_lora_path("model:sql-expert", None)
|
||||
self.assertEqual(result, "sql-expert")
|
||||
|
||||
def test_adapter_in_explicit_only(self):
|
||||
"""Test adapter from explicit lora_path when not in model."""
|
||||
result = self.serving._resolve_lora_path("model-name", "python-expert")
|
||||
self.assertEqual(result, "python-expert")
|
||||
|
||||
def test_model_parameter_takes_precedence(self):
|
||||
"""Test model parameter adapter takes precedence over explicit."""
|
||||
result = self.serving._resolve_lora_path("model:sql-expert", "python-expert")
|
||||
self.assertEqual(result, "sql-expert")
|
||||
|
||||
def test_with_list_explicit_lora_path(self):
|
||||
"""Test that explicit list is returned when no model adapter."""
|
||||
explicit = ["adapter1", "adapter2", None]
|
||||
result = self.serving._resolve_lora_path("model-name", explicit)
|
||||
self.assertEqual(result, explicit)
|
||||
|
||||
def test_model_adapter_overrides_list(self):
|
||||
"""Test model adapter overrides even when explicit is a list."""
|
||||
result = self.serving._resolve_lora_path(
|
||||
"model:sql-expert", ["adapter1", "adapter2"]
|
||||
)
|
||||
self.assertEqual(result, "sql-expert")
|
||||
|
||||
def test_complex_model_name_with_adapter(self):
|
||||
"""Test resolution with complex model name."""
|
||||
result = self.serving._resolve_lora_path(
|
||||
"org/model-v2.1:adapter-name", "other-adapter"
|
||||
)
|
||||
self.assertEqual(result, "adapter-name")
|
||||
|
||||
|
||||
class TestValidateLoraEnabled(unittest.TestCase):
|
||||
"""Test _validate_lora_enabled method."""
|
||||
|
||||
def test_validation_passes_when_lora_enabled(self):
|
||||
"""Test validation passes when LoRA is enabled."""
|
||||
tokenizer_manager = MockTokenizerManager(enable_lora=True)
|
||||
serving = ConcreteServingBase(tokenizer_manager)
|
||||
|
||||
# Should not raise
|
||||
try:
|
||||
serving._validate_lora_enabled("sql-expert")
|
||||
except ValueError:
|
||||
self.fail("_validate_lora_enabled raised ValueError unexpectedly")
|
||||
|
||||
def test_validation_fails_when_lora_disabled(self):
|
||||
"""Test validation fails with helpful message when LoRA is disabled."""
|
||||
tokenizer_manager = MockTokenizerManager(enable_lora=False)
|
||||
serving = ConcreteServingBase(tokenizer_manager)
|
||||
|
||||
with self.assertRaises(ValueError) as context:
|
||||
serving._validate_lora_enabled("sql-expert")
|
||||
|
||||
error_message = str(context.exception)
|
||||
self.assertIn("sql-expert", error_message)
|
||||
self.assertIn("--enable-lora", error_message)
|
||||
self.assertIn("not enabled", error_message)
|
||||
|
||||
def test_validation_error_mentions_adapter_name(self):
|
||||
"""Test that error message includes the requested adapter name."""
|
||||
tokenizer_manager = MockTokenizerManager(enable_lora=False)
|
||||
serving = ConcreteServingBase(tokenizer_manager)
|
||||
|
||||
with self.assertRaises(ValueError) as context:
|
||||
serving._validate_lora_enabled("my-custom-adapter")
|
||||
|
||||
self.assertIn("my-custom-adapter", str(context.exception))
|
||||
|
||||
|
||||
class TestIntegrationScenarios(unittest.TestCase):
|
||||
"""Integration tests for common usage scenarios."""
|
||||
|
||||
def setUp(self):
|
||||
self.tokenizer_manager = MockTokenizerManager(enable_lora=True)
|
||||
self.serving = ConcreteServingBase(self.tokenizer_manager)
|
||||
|
||||
def test_openai_compatible_usage(self):
|
||||
"""Test typical OpenAI-compatible usage pattern."""
|
||||
# User specifies adapter in model parameter
|
||||
model = "meta-llama/Llama-3.1-8B:sql-expert"
|
||||
explicit_lora = None
|
||||
|
||||
lora_path = self.serving._resolve_lora_path(model, explicit_lora)
|
||||
self.assertEqual(lora_path, "sql-expert")
|
||||
|
||||
# Validation should pass
|
||||
self.serving._validate_lora_enabled(lora_path)
|
||||
|
||||
def test_backward_compatible_usage(self):
|
||||
"""Test backward-compatible usage with explicit lora_path."""
|
||||
model = "meta-llama/Llama-3.1-8B"
|
||||
explicit_lora = "sql-expert"
|
||||
|
||||
lora_path = self.serving._resolve_lora_path(model, explicit_lora)
|
||||
self.assertEqual(lora_path, "sql-expert")
|
||||
|
||||
# Validation should pass
|
||||
self.serving._validate_lora_enabled(lora_path)
|
||||
|
||||
def test_base_model_usage(self):
|
||||
"""Test using base model without any adapter."""
|
||||
model = "meta-llama/Llama-3.1-8B"
|
||||
explicit_lora = None
|
||||
|
||||
lora_path = self.serving._resolve_lora_path(model, explicit_lora)
|
||||
self.assertIsNone(lora_path)
|
||||
|
||||
# No validation needed when no adapter
|
||||
|
||||
def test_batch_request_scenario(self):
|
||||
"""Test batch request with list of adapters."""
|
||||
model = "meta-llama/Llama-3.1-8B" # No adapter in model
|
||||
explicit_lora = ["sql-expert", "python-expert", None]
|
||||
|
||||
lora_path = self.serving._resolve_lora_path(model, explicit_lora)
|
||||
self.assertEqual(lora_path, explicit_lora)
|
||||
|
||||
# Validate first adapter in list
|
||||
if isinstance(lora_path, list) and lora_path[0]:
|
||||
self.serving._validate_lora_enabled(lora_path[0])
|
||||
|
||||
def test_adapter_in_model_overrides_batch_list(self):
|
||||
"""Test that adapter in model parameter overrides batch list."""
|
||||
model = "meta-llama/Llama-3.1-8B:preferred-adapter"
|
||||
explicit_lora = ["adapter1", "adapter2"]
|
||||
|
||||
lora_path = self.serving._resolve_lora_path(model, explicit_lora)
|
||||
self.assertEqual(lora_path, "preferred-adapter")
|
||||
|
||||
def test_error_when_lora_not_enabled(self):
|
||||
"""Test comprehensive error flow when LoRA is not enabled."""
|
||||
# Setup server without LoRA enabled
|
||||
tokenizer_manager = MockTokenizerManager(enable_lora=False)
|
||||
serving = ConcreteServingBase(tokenizer_manager)
|
||||
|
||||
# User tries to use adapter
|
||||
model = "meta-llama/Llama-3.1-8B:sql-expert"
|
||||
lora_path = serving._resolve_lora_path(model, None)
|
||||
|
||||
# Should get helpful error
|
||||
with self.assertRaises(ValueError) as context:
|
||||
serving._validate_lora_enabled(lora_path)
|
||||
|
||||
error = str(context.exception)
|
||||
self.assertIn("--enable-lora", error)
|
||||
self.assertIn("sql-expert", error)
|
||||
|
||||
|
||||
class TestEdgeCases(unittest.TestCase):
|
||||
"""Test edge cases and error conditions."""
|
||||
|
||||
def setUp(self):
|
||||
self.tokenizer_manager = MockTokenizerManager(enable_lora=True)
|
||||
self.serving = ConcreteServingBase(self.tokenizer_manager)
|
||||
|
||||
def test_empty_string_model(self):
|
||||
"""Test handling of empty string model."""
|
||||
base, adapter = self.serving._parse_model_parameter("")
|
||||
self.assertEqual(base, "")
|
||||
self.assertIsNone(adapter)
|
||||
|
||||
def test_only_colon(self):
|
||||
"""Test model parameter that is just a colon."""
|
||||
base, adapter = self.serving._parse_model_parameter(":")
|
||||
self.assertEqual(base, "")
|
||||
self.assertIsNone(adapter)
|
||||
|
||||
def test_empty_list_lora_path(self):
|
||||
"""Test validation with empty list doesn't crash."""
|
||||
lora_path = self.serving._resolve_lora_path("model-name", [])
|
||||
# Empty list is falsy, so validation won't be called
|
||||
self.assertEqual(lora_path, [])
|
||||
|
||||
def test_list_with_none_first(self):
|
||||
"""Test validation finds first non-None adapter in list."""
|
||||
lora_path = self.serving._resolve_lora_path("model-name", [None, "adapter2"])
|
||||
self.assertEqual(lora_path, [None, "adapter2"])
|
||||
# In actual usage, validation would find "adapter2"
|
||||
|
||||
def test_list_all_none(self):
|
||||
"""Test validation with list of all None values."""
|
||||
lora_path = self.serving._resolve_lora_path("model-name", [None, None])
|
||||
self.assertEqual(lora_path, [None, None])
|
||||
# In actual usage, no validation would occur (no non-None adapters)
|
||||
|
||||
def test_unicode_in_adapter_name(self):
|
||||
"""Test Unicode characters in adapter name."""
|
||||
base, adapter = self.serving._parse_model_parameter("model:adapter-名前")
|
||||
self.assertEqual(base, "model")
|
||||
self.assertEqual(adapter, "adapter-名前")
|
||||
|
||||
def test_special_characters_in_adapter(self):
|
||||
"""Test special characters in adapter name."""
|
||||
base, adapter = self.serving._parse_model_parameter("model:adapter_v2.1-final")
|
||||
self.assertEqual(base, "model")
|
||||
self.assertEqual(adapter, "adapter_v2.1-final")
|
||||
|
||||
def test_none_as_explicit_lora_path(self):
|
||||
"""Test None as explicit lora_path is handled correctly."""
|
||||
result = self.serving._resolve_lora_path("model:adapter", None)
|
||||
self.assertEqual(result, "adapter")
|
||||
|
||||
def test_empty_string_as_explicit_lora_path(self):
|
||||
"""Test empty string as explicit lora_path."""
|
||||
result = self.serving._resolve_lora_path("model-name", "")
|
||||
self.assertEqual(result, "")
|
||||
|
||||
def test_validation_with_empty_adapter_name(self):
|
||||
"""Test validation with empty adapter name still raises error."""
|
||||
tokenizer_manager = MockTokenizerManager(enable_lora=False)
|
||||
serving = ConcreteServingBase(tokenizer_manager)
|
||||
|
||||
with self.assertRaises(ValueError):
|
||||
serving._validate_lora_enabled("")
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
@@ -1,282 +0,0 @@
|
||||
"""
|
||||
End-to-end tests for OpenAI-compatible LoRA adapter usage.
|
||||
|
||||
Tests the model:adapter syntax and backward compatibility with explicit lora_path.
|
||||
|
||||
Usage:
|
||||
python3 -m unittest openai_server.features.test_lora_openai_compatible.TestLoRAOpenAICompatible.test_model_adapter_syntax
|
||||
python3 -m unittest openai_server.features.test_lora_openai_compatible.TestLoRAOpenAICompatible.test_explicit_lora_path
|
||||
python3 -m unittest openai_server.features.test_lora_openai_compatible.TestLoRAOpenAICompatible.test_priority_model_over_explicit
|
||||
python3 -m unittest openai_server.features.test_lora_openai_compatible.TestLoRAOpenAICompatible.test_base_model_no_adapter
|
||||
python3 -m unittest openai_server.features.test_lora_openai_compatible.TestLoRAOpenAICompatible.test_completions_api_with_adapter
|
||||
python3 -m unittest openai_server.features.test_lora_openai_compatible.TestLoRAOpenAICompatible.test_streaming_with_adapter
|
||||
python3 -m unittest openai_server.features.test_lora_openai_compatible.TestLoRADisabledError.test_lora_disabled_error
|
||||
"""
|
||||
|
||||
import unittest
|
||||
|
||||
import openai
|
||||
|
||||
from sglang.test.ci.ci_register import register_cuda_ci
|
||||
|
||||
register_cuda_ci(est_time=150, suite="nightly-1-gpu", nightly=True)
|
||||
|
||||
from sglang.srt.utils import kill_process_tree
|
||||
from sglang.test.test_utils import (
|
||||
DEFAULT_SMALL_MODEL_NAME_FOR_TEST,
|
||||
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
||||
DEFAULT_URL_FOR_TEST,
|
||||
CustomTestCase,
|
||||
popen_launch_server,
|
||||
)
|
||||
|
||||
|
||||
def get_real_lora_adapter() -> str:
|
||||
"""Use a real LoRA adapter from Hugging Face."""
|
||||
return "codelion/Llama-3.2-1B-Instruct-tool-calling-lora"
|
||||
|
||||
|
||||
def setup_class(cls, enable_lora=True):
|
||||
"""Setup test class with LoRA-enabled server."""
|
||||
cls.model = DEFAULT_SMALL_MODEL_NAME_FOR_TEST
|
||||
cls.base_url = DEFAULT_URL_FOR_TEST
|
||||
|
||||
# Use real LoRA adapter
|
||||
cls.lora_adapter_path = get_real_lora_adapter()
|
||||
|
||||
other_args = [
|
||||
"--max-running-requests",
|
||||
"10",
|
||||
"--disable-radix-cache", # Disable cache for cleaner tests
|
||||
]
|
||||
|
||||
if enable_lora:
|
||||
other_args.extend(
|
||||
[
|
||||
"--enable-lora",
|
||||
"--lora-paths",
|
||||
f"tool_calling={cls.lora_adapter_path}",
|
||||
]
|
||||
)
|
||||
|
||||
cls.process = popen_launch_server(
|
||||
cls.model,
|
||||
cls.base_url,
|
||||
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
||||
other_args=other_args,
|
||||
)
|
||||
cls.client = openai.Client(api_key="EMPTY", base_url=f"{cls.base_url}/v1")
|
||||
|
||||
|
||||
class TestLoRAOpenAICompatible(CustomTestCase):
|
||||
"""Test OpenAI-compatible LoRA adapter usage."""
|
||||
|
||||
@classmethod
|
||||
def setUpClass(cls):
|
||||
setup_class(cls, enable_lora=True)
|
||||
|
||||
@classmethod
|
||||
def tearDownClass(cls):
|
||||
kill_process_tree(cls.process.pid)
|
||||
|
||||
def test_model_adapter_syntax(self):
|
||||
"""Test the new model:adapter syntax works correctly."""
|
||||
response = self.client.chat.completions.create(
|
||||
# ← New OpenAI-compatible syntax
|
||||
model=f"{self.model}:tool_calling",
|
||||
messages=[{"role": "user", "content": "What tools do you have available?"}],
|
||||
max_tokens=50,
|
||||
temperature=0,
|
||||
)
|
||||
|
||||
self.assertIsNotNone(response.choices[0].message.content)
|
||||
self.assertGreater(len(response.choices[0].message.content), 0)
|
||||
print(f"Model adapter syntax response: {response.choices[0].message.content}")
|
||||
|
||||
def test_explicit_lora_path(self):
|
||||
"""Test backward compatibility with explicit lora_path via extra_body."""
|
||||
response = self.client.chat.completions.create(
|
||||
model=self.model,
|
||||
messages=[{"role": "user", "content": "What tools do you have available?"}],
|
||||
# ← Legacy explicit method
|
||||
extra_body={"lora_path": "tool_calling"},
|
||||
max_tokens=50,
|
||||
temperature=0,
|
||||
)
|
||||
|
||||
self.assertIsNotNone(response.choices[0].message.content)
|
||||
self.assertGreater(len(response.choices[0].message.content), 0)
|
||||
print(f"Explicit lora_path response: {response.choices[0].message.content}")
|
||||
|
||||
def test_priority_model_over_explicit(self):
|
||||
"""Test that model:adapter syntax takes precedence over explicit lora_path."""
|
||||
# This test verifies the priority logic in _resolve_lora_path
|
||||
response = self.client.chat.completions.create(
|
||||
# ← Model specifies tool_calling adapter
|
||||
model=f"{self.model}:tool_calling",
|
||||
messages=[{"role": "user", "content": "What tools do you have available?"}],
|
||||
# ← Both specify same adapter
|
||||
extra_body={"lora_path": "tool_calling"},
|
||||
max_tokens=50,
|
||||
temperature=0,
|
||||
)
|
||||
|
||||
# Should use tool_calling adapter (model parameter takes precedence)
|
||||
self.assertIsNotNone(response.choices[0].message.content)
|
||||
self.assertGreater(len(response.choices[0].message.content), 0)
|
||||
print(f"Priority test response: {response.choices[0].message.content}")
|
||||
|
||||
def test_base_model_no_adapter(self):
|
||||
"""Test using base model without any adapter."""
|
||||
response = self.client.chat.completions.create(
|
||||
model=self.model, # ← No adapter specified
|
||||
messages=[{"role": "user", "content": "Hello, how are you?"}],
|
||||
max_tokens=30,
|
||||
temperature=0,
|
||||
)
|
||||
|
||||
self.assertIsNotNone(response.choices[0].message.content)
|
||||
self.assertGreater(len(response.choices[0].message.content), 0)
|
||||
print(f"Base model response: {response.choices[0].message.content}")
|
||||
|
||||
def test_completions_api_with_adapter(self):
|
||||
"""Test completions API with LoRA adapter."""
|
||||
response = self.client.completions.create(
|
||||
model=f"{self.model}:tool_calling", # ← Using model:adapter syntax
|
||||
prompt="What tools do you have available?",
|
||||
max_tokens=50,
|
||||
temperature=0,
|
||||
)
|
||||
|
||||
self.assertIsNotNone(response.choices[0].text)
|
||||
self.assertGreater(len(response.choices[0].text), 0)
|
||||
print(f"Completions API response: {response.choices[0].text}")
|
||||
|
||||
def test_streaming_with_adapter(self):
|
||||
"""Test streaming with LoRA adapter."""
|
||||
stream = self.client.chat.completions.create(
|
||||
model=f"{self.model}:tool_calling",
|
||||
messages=[{"role": "user", "content": "What tools do you have available?"}],
|
||||
max_tokens=50,
|
||||
temperature=0,
|
||||
stream=True,
|
||||
)
|
||||
|
||||
collected_content = ""
|
||||
for chunk in stream:
|
||||
if chunk.choices[0].delta.content:
|
||||
collected_content += chunk.choices[0].delta.content
|
||||
|
||||
self.assertGreater(len(collected_content), 0)
|
||||
print(f"Streaming response: {collected_content}")
|
||||
|
||||
def test_multiple_adapters(self):
|
||||
"""Test using different adapters in sequence."""
|
||||
# Test tool_calling adapter
|
||||
tool_response = self.client.chat.completions.create(
|
||||
model=f"{self.model}:tool_calling",
|
||||
messages=[{"role": "user", "content": "What tools do you have available?"}],
|
||||
max_tokens=30,
|
||||
temperature=0,
|
||||
)
|
||||
|
||||
# Test base model without adapter
|
||||
base_response = self.client.chat.completions.create(
|
||||
model=self.model,
|
||||
messages=[{"role": "user", "content": "Hello, how are you?"}],
|
||||
max_tokens=30,
|
||||
temperature=0,
|
||||
)
|
||||
|
||||
self.assertIsNotNone(tool_response.choices[0].message.content)
|
||||
self.assertIsNotNone(base_response.choices[0].message.content)
|
||||
print(
|
||||
f"Tool calling adapter response: {tool_response.choices[0].message.content}"
|
||||
)
|
||||
print(f"Base model response: {base_response.choices[0].message.content}")
|
||||
|
||||
|
||||
class TestLoRADisabledError(CustomTestCase):
|
||||
"""Test error handling when LoRA is disabled."""
|
||||
|
||||
@classmethod
|
||||
def setUpClass(cls):
|
||||
setup_class(cls, enable_lora=False) # ← LoRA disabled
|
||||
|
||||
@classmethod
|
||||
def tearDownClass(cls):
|
||||
kill_process_tree(cls.process.pid)
|
||||
|
||||
def test_lora_disabled_error(self):
|
||||
"""Test that using LoRA adapter when LoRA is disabled raises appropriate error."""
|
||||
with self.assertRaises(openai.APIError) as context:
|
||||
self.client.chat.completions.create(
|
||||
model=f"{self.model}:tool_calling", # ← Trying to use adapter
|
||||
messages=[
|
||||
{"role": "user", "content": "What tools do you have available?"}
|
||||
],
|
||||
max_tokens=50,
|
||||
)
|
||||
|
||||
# Verify the error message contains helpful guidance
|
||||
error_message = str(context.exception)
|
||||
self.assertIn("LoRA", error_message)
|
||||
self.assertIn("not enabled", error_message)
|
||||
print(f"Expected error message: {error_message}")
|
||||
|
||||
|
||||
class TestLoRAEdgeCases(CustomTestCase):
|
||||
"""Test edge cases for LoRA adapter usage."""
|
||||
|
||||
@classmethod
|
||||
def setUpClass(cls):
|
||||
setup_class(cls, enable_lora=True)
|
||||
|
||||
@classmethod
|
||||
def tearDownClass(cls):
|
||||
kill_process_tree(cls.process.pid)
|
||||
|
||||
def test_model_with_colon_no_adapter(self):
|
||||
"""Test model parameter ending with colon (empty adapter)."""
|
||||
response = self.client.chat.completions.create(
|
||||
model=f"{self.model}:", # ← Model ends with colon
|
||||
messages=[{"role": "user", "content": "Hello!"}],
|
||||
max_tokens=30,
|
||||
temperature=0,
|
||||
)
|
||||
|
||||
# Should work as base model (no adapter)
|
||||
self.assertIsNotNone(response.choices[0].message.content)
|
||||
print(f"Model with colon response: {response.choices[0].message.content}")
|
||||
|
||||
def test_explicit_lora_path_none(self):
|
||||
"""Test explicit lora_path set to None."""
|
||||
response = self.client.chat.completions.create(
|
||||
model=self.model,
|
||||
messages=[{"role": "user", "content": "Hello!"}],
|
||||
extra_body={"lora_path": None}, # ← Explicitly None
|
||||
max_tokens=30,
|
||||
temperature=0,
|
||||
)
|
||||
|
||||
# Should work as base model
|
||||
self.assertIsNotNone(response.choices[0].message.content)
|
||||
print(
|
||||
f"Explicit None lora_path response: {response.choices[0].message.content}"
|
||||
)
|
||||
|
||||
def test_invalid_adapter_name(self):
|
||||
"""Test using non-existent adapter name."""
|
||||
with self.assertRaises(openai.APIError) as context:
|
||||
self.client.chat.completions.create(
|
||||
model=f"{self.model}:nonexistent", # ← Non-existent adapter
|
||||
messages=[{"role": "user", "content": "Hello!"}],
|
||||
max_tokens=30,
|
||||
)
|
||||
|
||||
error_message = str(context.exception)
|
||||
print(f"Invalid adapter error: {error_message}")
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
@@ -1,65 +0,0 @@
|
||||
# Copyright 2023-2025 SGLang Team
|
||||
# Licensed under the Apache License, Version 2.0 (the "License");
|
||||
# you may not use this file except in compliance with the License.
|
||||
# You may obtain a copy of the License at
|
||||
#
|
||||
# http://www.apache.org/licenses/LICENSE-2.0
|
||||
#
|
||||
# Unless required by applicable law or agreed to in writing, software
|
||||
# distributed under the License is distributed on an "AS IS" BASIS,
|
||||
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||
# See the License for the specific language governing permissions and
|
||||
# limitations under the License.
|
||||
# ==============================================================================
|
||||
|
||||
import multiprocessing as mp
|
||||
import sys
|
||||
import unittest
|
||||
from pathlib import Path
|
||||
|
||||
# Add test directory to path for lora_utils import
|
||||
# TODO: can be removed after migration
|
||||
sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
|
||||
|
||||
from lora_utils import (
|
||||
LoRAAdaptor,
|
||||
LoRAModelCase,
|
||||
run_lora_multiple_batch_on_model_cases,
|
||||
)
|
||||
|
||||
from sglang.test.ci.ci_register import register_cuda_ci
|
||||
|
||||
register_cuda_ci(est_time=97, suite="nightly-1-gpu", nightly=True)
|
||||
|
||||
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 TestLoRAQwen3(CustomTestCase):
|
||||
def test_ci_lora_models(self):
|
||||
run_lora_multiple_batch_on_model_cases(LORA_MODELS_QWEN3)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
try:
|
||||
mp.set_start_method("spawn")
|
||||
except RuntimeError:
|
||||
pass
|
||||
|
||||
unittest.main(warnings="ignore")
|
||||
@@ -1,89 +0,0 @@
|
||||
# Copyright 2023-2024 SGLang Team
|
||||
# Licensed under the Apache License, Version 2.0 (the "License");
|
||||
# you may not use this file except in compliance with the License.
|
||||
# You may obtain a copy of the License at
|
||||
#
|
||||
# http://www.apache.org/licenses/LICENSE-2.0
|
||||
#
|
||||
# Unless required by applicable law or agreed to in writing, software
|
||||
# distributed under the License is distributed on an "AS IS" BASIS,
|
||||
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||
# See the License for the specific language governing permissions and
|
||||
# limitations under the License.
|
||||
# ==============================================================================
|
||||
|
||||
import multiprocessing as mp
|
||||
import sys
|
||||
import unittest
|
||||
from pathlib import Path
|
||||
|
||||
import torch
|
||||
|
||||
# Add test directory to path for lora_utils import
|
||||
# TODO: can be removed after migration
|
||||
sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
|
||||
|
||||
from lora_utils import CI_MULTI_LORA_MODELS, run_lora_test_one_by_one
|
||||
|
||||
from sglang.test.ci.ci_register import register_cuda_ci
|
||||
|
||||
register_cuda_ci(est_time=200, suite="nightly-1-gpu", nightly=True)
|
||||
|
||||
from sglang.test.test_utils import CustomTestCase
|
||||
|
||||
PROMPTS = [
|
||||
"AI is a field of computer science focused on",
|
||||
"""
|
||||
### Instruction:
|
||||
Tell me about llamas and alpacas
|
||||
### Response:
|
||||
Llamas are large, long-necked animals with a woolly coat. They have two toes on each foot instead of three like other camelids.
|
||||
### Question:
|
||||
What do you know about llamas?
|
||||
### Answer:
|
||||
""",
|
||||
]
|
||||
|
||||
|
||||
class TestLoRARadixCache(CustomTestCase):
|
||||
|
||||
def test_lora_radix_cache(self):
|
||||
# Here we need a model case with multiple adaptors for testing correctness of radix cache
|
||||
model_case = CI_MULTI_LORA_MODELS[0]
|
||||
|
||||
torch_dtype = torch.float16
|
||||
max_new_tokens = 32
|
||||
batch_prompts = (
|
||||
PROMPTS
|
||||
if not model_case.skip_long_prompt
|
||||
else [p for p in PROMPTS if len(p) < 1000]
|
||||
)
|
||||
|
||||
# Test lora with radix cache
|
||||
run_lora_test_one_by_one(
|
||||
batch_prompts,
|
||||
model_case,
|
||||
torch_dtype,
|
||||
max_new_tokens=max_new_tokens,
|
||||
disable_radix_cache=False,
|
||||
test_tag="lora-with-radix-cache",
|
||||
)
|
||||
|
||||
# Test lora without radix cache
|
||||
run_lora_test_one_by_one(
|
||||
batch_prompts,
|
||||
model_case,
|
||||
torch_dtype,
|
||||
max_new_tokens=max_new_tokens,
|
||||
disable_radix_cache=True,
|
||||
test_tag="lora-without-radix-cache",
|
||||
)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
try:
|
||||
mp.set_start_method("spawn")
|
||||
except RuntimeError:
|
||||
pass
|
||||
|
||||
unittest.main(warnings="ignore")
|
||||
Reference in New Issue
Block a user