Skip to content

GitLab

Commit cc7f22a8 authored by zhuwenwen's avatar zhuwenwen
Browse files

Merge tag 'v0.9.1' into v0.9.1-ori

parents b9ea0c09 b6553be1
Show whitespace changes
Inline Side-by-side
Showing with 60 additions and 56 deletions
tests/models/quantization/test_nvfp4.py
View file @ cc7f22a8
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
# flake8: noqa
"""Tests Model Optimizer nvfp4 models against ground truth generation
Note: these tests will only pass on B200
......
tests/models/registry.py
View file @ cc7f22a8
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
from collections.abc import Mapping, Set
from dataclasses import dataclass, field
......@@ -159,17 +160,12 @@ _TEXT_GENERATION_EXAMPLE_MODELS = {
"Fairseq2LlamaForCausalLM": _HfExamplesInfo("mgleize/fairseq2-dummy-Llama-3.2-1B"), # noqa: E501
"FalconForCausalLM": _HfExamplesInfo("tiiuae/falcon-7b"),
"FalconH1ForCausalLM":_HfExamplesInfo("tiiuae/Falcon-H1-1.5B-Instruct",
is_available_online=False,
min_transformers_version="4.52.2"),
min_transformers_version="4.53"),
"GemmaForCausalLM": _HfExamplesInfo("google/gemma-1.1-2b-it"),
"Gemma2ForCausalLM": _HfExamplesInfo("google/gemma-2-9b"),
"Gemma3ForCausalLM": _HfExamplesInfo("google/gemma-3-1b-it"),
"GlmForCausalLM": _HfExamplesInfo("THUDM/glm-4-9b-chat-hf"),
"Glm4ForCausalLM": _HfExamplesInfo(
"THUDM/GLM-4-32B-0414",
is_available_online=False,
min_transformers_version="4.52.dev0"
),
"Glm4ForCausalLM": _HfExamplesInfo("THUDM/GLM-4-9B-0414"),
"GPT2LMHeadModel": _HfExamplesInfo("openai-community/gpt2",
{"alias": "gpt2"}),
"GPTBigCodeForCausalLM": _HfExamplesInfo("bigcode/starcoder",
......@@ -180,8 +176,7 @@ _TEXT_GENERATION_EXAMPLE_MODELS = {
{"1b": "EleutherAI/pythia-1.4b"}),
"GraniteForCausalLM": _HfExamplesInfo("ibm/PowerLM-3b"),
"GraniteMoeForCausalLM": _HfExamplesInfo("ibm/PowerMoE-3b"),
"GraniteMoeHybridForCausalLM": _HfExamplesInfo("ibm-granite/granite-4.0-tiny-preview", # noqa: E501
min_transformers_version="4.52.0"), # noqa: E501
"GraniteMoeHybridForCausalLM": _HfExamplesInfo("ibm-granite/granite-4.0-tiny-preview"), # noqa: E501
"GraniteMoeSharedForCausalLM": _HfExamplesInfo("ibm-research/moe-7b-1b-active-shared-experts"), # noqa: E501
"Grok1ModelForCausalLM": _HfExamplesInfo("hpcai-tech/grok-1",
trust_remote_code=True),
......@@ -202,8 +197,7 @@ _TEXT_GENERATION_EXAMPLE_MODELS = {
"LLaMAForCausalLM": _HfExamplesInfo("decapoda-research/llama-7b-hf",
is_available_online=False),
"MambaForCausalLM": _HfExamplesInfo("state-spaces/mamba-130m-hf"),
"Mamba2ForCausalLM": _HfExamplesInfo("mistralai/Mamba-Codestral-7B-v0.1",
is_available_online=False),
"Mamba2ForCausalLM": _HfExamplesInfo("mistralai/Mamba-Codestral-7B-v0.1"),
"FalconMambaForCausalLM": _HfExamplesInfo("tiiuae/falcon-mamba-7b-instruct"), # noqa: E501
"MiniCPMForCausalLM": _HfExamplesInfo("openbmb/MiniCPM-2B-sft-bf16",
trust_remote_code=True),
......@@ -218,6 +212,8 @@ _TEXT_GENERATION_EXAMPLE_MODELS = {
"MptForCausalLM": _HfExamplesInfo("mpt", is_available_online=False),
"MPTForCausalLM": _HfExamplesInfo("mosaicml/mpt-7b"),
"NemotronForCausalLM": _HfExamplesInfo("nvidia/Minitron-8B-Base"),
"NemotronHForCausalLM": _HfExamplesInfo("nvidia/Nemotron-H-8B-Base-8K",
trust_remote_code=True),
"OlmoForCausalLM": _HfExamplesInfo("allenai/OLMo-1B-hf"),
"Olmo2ForCausalLM": _HfExamplesInfo("allenai/OLMo-2-0425-1B"),
"OlmoeForCausalLM": _HfExamplesInfo("allenai/OLMoE-1B-7B-0924-Instruct"),
......@@ -242,10 +238,9 @@ _TEXT_GENERATION_EXAMPLE_MODELS = {
"Qwen2MoeForCausalLM": _HfExamplesInfo("Qwen/Qwen1.5-MoE-A2.7B-Chat"),
"Qwen3ForCausalLM": _HfExamplesInfo("Qwen/Qwen3-8B"),
"Qwen3MoeForCausalLM": _HfExamplesInfo("Qwen/Qwen3-30B-A3B"),
"RWForCausalLM": _HfExamplesInfo("tiiuae/falcon-40b",
is_available_online=False),
"RWForCausalLM": _HfExamplesInfo("tiiuae/falcon-40b"),
"StableLMEpochForCausalLM": _HfExamplesInfo("stabilityai/stablelm-zephyr-3b", # noqa: E501
is_available_online=False),
v0_only=True),
"StableLmForCausalLM": _HfExamplesInfo("stabilityai/stablelm-3b-4e1t",
v0_only=True),
"Starcoder2ForCausalLM": _HfExamplesInfo("bigcode/starcoder2-3b"),
......@@ -255,7 +250,7 @@ _TEXT_GENERATION_EXAMPLE_MODELS = {
"TeleFLMForCausalLM": _HfExamplesInfo("CofeAI/FLM-2-52B-Instruct-2407",
trust_remote_code=True),
"XverseForCausalLM": _HfExamplesInfo("xverse/XVERSE-7B-Chat",
is_available_online=False,
tokenizer="meta-llama/Llama-2-7b",
trust_remote_code=True),
"Zamba2ForCausalLM": _HfExamplesInfo("Zyphra/Zamba2-7B-instruct"),
"MiMoForCausalLM": _HfExamplesInfo("XiaomiMiMo/MiMo-7B-RL",
......@@ -274,8 +269,7 @@ _EMBEDDING_EXAMPLE_MODELS = {
trust_remote_code=True),
"GteNewModel": _HfExamplesInfo("Alibaba-NLP/gte-base-en-v1.5",
trust_remote_code=True,
hf_overrides={"architectures":
["GteNewModel"]}),
hf_overrides={"architectures": ["GteNewModel"]}), # noqa: E501
"InternLM2ForRewardModel": _HfExamplesInfo("internlm/internlm2-1_8b-reward",
trust_remote_code=True),
"JambaForSequenceClassification": _HfExamplesInfo("ai21labs/Jamba-tiny-reward-dev"), # noqa: E501
......@@ -283,7 +277,7 @@ _EMBEDDING_EXAMPLE_MODELS = {
"MistralModel": _HfExamplesInfo("intfloat/e5-mistral-7b-instruct"),
"ModernBertModel": _HfExamplesInfo("Alibaba-NLP/gte-modernbert-base",
trust_remote_code=True),
"NomicBertModel": _HfExamplesInfo("Snowflake/snowflake-arctic-embed-m-long", # noqa: E501
"NomicBertModel": _HfExamplesInfo("nomic-ai/nomic-embed-text-v2-moe",
trust_remote_code=True),
"Qwen2Model": _HfExamplesInfo("ssmits/Qwen2-7B-Instruct-embed-base"),
"Qwen2ForRewardModel": _HfExamplesInfo("Qwen/Qwen2.5-Math-RM-72B"),
......@@ -297,10 +291,8 @@ _EMBEDDING_EXAMPLE_MODELS = {
"Phi3VForCausalLM": _HfExamplesInfo("TIGER-Lab/VLM2Vec-Full",
trust_remote_code=True),
"Qwen2VLForConditionalGeneration": _HfExamplesInfo("MrLight/dse-qwen2-2b-mrl-v1"), # noqa: E501
# The model on Huggingface is currently being updated,
# hence I temporarily mark it as not available online
"PrithviGeoSpatialMAE": _HfExamplesInfo("ibm-nasa-geospatial/Prithvi-EO-2.0-300M-TL-Sen1Floods11", # noqa: E501
is_available_online=False),
is_available_online=False), # noqa: E501
}
_CROSS_ENCODER_EXAMPLE_MODELS = {
......@@ -326,8 +318,7 @@ _MULTIMODAL_EXAMPLE_MODELS = {
hf_overrides={"architectures": ["DeepseekVLV2ForCausalLM"]}), # noqa: E501
"FuyuForCausalLM": _HfExamplesInfo("adept/fuyu-8b"),
"Gemma3ForConditionalGeneration": _HfExamplesInfo("google/gemma-3-4b-it"),
"GraniteSpeechForConditionalGeneration": _HfExamplesInfo("ibm-granite/granite-speech-3.3-8b", # noqa: E501
min_transformers_version="4.52.0"), # noqa: E501
"GraniteSpeechForConditionalGeneration": _HfExamplesInfo("ibm-granite/granite-speech-3.3-2b"), # noqa: E501
"GLM4VForCausalLM": _HfExamplesInfo("THUDM/glm-4v-9b",
trust_remote_code=True,
hf_overrides={"architectures": ["GLM4VForCausalLM"]}), # noqa: E501
......@@ -346,7 +337,6 @@ _MULTIMODAL_EXAMPLE_MODELS = {
trust_remote_code=True,
v0_only=True),
"Llama4ForConditionalGeneration": _HfExamplesInfo("meta-llama/Llama-4-Scout-17B-16E-Instruct", # noqa: E501
min_transformers_version="4.51",
max_model_len=10240),
"LlavaForConditionalGeneration": _HfExamplesInfo("llava-hf/llava-1.5-7b-hf",
extras={"mistral": "mistral-community/pixtral-12b", # noqa: E501
......@@ -359,8 +349,6 @@ _MULTIMODAL_EXAMPLE_MODELS = {
transformers_version_reason="HF model is not compatible.", # noqa: E501
hf_overrides={"architectures": ["MantisForConditionalGeneration"]}), # noqa: E501
"MiniCPMO": _HfExamplesInfo("openbmb/MiniCPM-o-2_6",
max_transformers_version="4.48",
transformers_version_reason="Use of deprecated imports which have been removed.", # noqa: E501
trust_remote_code=True),
"MiniCPMV": _HfExamplesInfo("openbmb/MiniCPM-Llama3-V-2_5",
extras={"2.6": "openbmb/MiniCPM-V-2_6"}, # noqa: E501
......@@ -398,20 +386,20 @@ _MULTIMODAL_EXAMPLE_MODELS = {
"Qwen2AudioForConditionalGeneration": _HfExamplesInfo("Qwen/Qwen2-Audio-7B-Instruct"), # noqa: E501
"Qwen2VLForConditionalGeneration": _HfExamplesInfo("Qwen/Qwen2-VL-2B-Instruct"), # noqa: E501
"Qwen2_5_VLForConditionalGeneration": _HfExamplesInfo("Qwen/Qwen2.5-VL-3B-Instruct"), # noqa: E501
"Qwen2_5OmniModel": _HfExamplesInfo("Qwen/Qwen2.5-Omni-3B",
min_transformers_version="4.52"),
"Qwen2_5OmniForConditionalGeneration": _HfExamplesInfo("Qwen/Qwen2.5-Omni-7B-AWQ", # noqa: E501
min_transformers_version="4.52"),
"Qwen2_5OmniModel": _HfExamplesInfo("Qwen/Qwen2.5-Omni-3B"),
"Qwen2_5OmniForConditionalGeneration": _HfExamplesInfo("Qwen/Qwen2.5-Omni-7B-AWQ"), # noqa: E501
"SkyworkR1VChatModel": _HfExamplesInfo("Skywork/Skywork-R1V-38B"),
"SmolVLMForConditionalGeneration": _HfExamplesInfo("HuggingFaceTB/SmolVLM2-2.2B-Instruct"), # noqa: E501
"UltravoxModel": _HfExamplesInfo("fixie-ai/ultravox-v0_5-llama-3_2-1b", # noqa: E501
trust_remote_code=True),
"TarsierForConditionalGeneration": _HfExamplesInfo("omni-research/Tarsier-7b", # noqa: E501
hf_overrides={"architectures": ["TarsierForConditionalGeneration"]}), # noqa: E501
# [Encoder-decoder]
# Florence-2 uses BartFastTokenizer which can't be loaded from AutoTokenizer
# Therefore, we borrow the BartTokenizer from the original Bart model
"Florence2ForConditionalGeneration": _HfExamplesInfo("microsoft/Florence-2-base", # noqa: E501
tokenizer="Isotr0py/Florence-2-tokenizer",
trust_remote_code=True,), # noqa: E501
tokenizer="Isotr0py/Florence-2-tokenizer", # noqa: E501
trust_remote_code=True), # noqa: E501
"MllamaForConditionalGeneration": _HfExamplesInfo("meta-llama/Llama-3.2-11B-Vision-Instruct"), # noqa: E501
"Llama4ForConditionalGeneration": _HfExamplesInfo("meta-llama/Llama-4-Scout-17B-16E-Instruct"), # noqa: E501
"WhisperForConditionalGeneration": _HfExamplesInfo("openai/whisper-large-v3"), # noqa: E501
......@@ -435,6 +423,11 @@ _SPECULATIVE_DECODING_EXAMPLE_MODELS = {
trust_remote_code=True,
speculative_model="yuhuili/EAGLE3-LLaMA3.1-Instruct-8B",
tokenizer="meta-llama/Llama-3.1-8B-Instruct"),
"EagleMiniCPMForCausalLM": _HfExamplesInfo("openbmb/MiniCPM-1B-sft-bf16",
trust_remote_code=True,
is_available_online=False,
speculative_model="openbmb/MiniCPM-2B-sft-bf16",
tokenizer="openbmb/MiniCPM-2B-sft-bf16"),
"MiMoMTPModel": _HfExamplesInfo("XiaomiMiMo/MiMo-7B-RL",
trust_remote_code=True,
speculative_model="XiaomiMiMo/MiMo-7B-RL")
......
tests/models/test_initialization.py
View file @ cc7f22a8
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
from unittest.mock import patch
......@@ -20,6 +21,10 @@ def test_can_initialize(model_arch: str, monkeypatch: pytest.MonkeyPatch):
model_info.check_available_online(on_fail="skip")
model_info.check_transformers_version(on_fail="skip")
# FIXME: Possible memory leak in the previous tests?
if model_arch == "GraniteSpeechForConditionalGeneration":
pytest.skip("Avoid OOM")
# Avoid OOM and reduce initialization time by only using 1 layer
def hf_overrides(hf_config: PretrainedConfig) -> PretrainedConfig:
hf_config.update(model_info.hf_overrides)
......@@ -40,6 +45,13 @@ def test_can_initialize(model_arch: str, monkeypatch: pytest.MonkeyPatch):
"num_hidden_layers": 1,
})
# e.g.: ibm-granite/granite-speech-3.3-2b
if hasattr(hf_config, "encoder_config"):
hf_config.encoder_config.update({
"num_layers": 1,
"num_hidden_layers": 1,
})
return hf_config
# Avoid calling model.forward()
......@@ -74,6 +86,8 @@ def test_can_initialize(model_arch: str, monkeypatch: pytest.MonkeyPatch):
} if model_info.speculative_model else None,
trust_remote_code=model_info.trust_remote_code,
max_model_len=model_info.max_model_len,
# these tests seem to produce leftover memory
gpu_memory_utilization=0.80,
load_format="dummy",
hf_overrides=hf_overrides,
)
tests/models/test_oot_registration.py
View file @ cc7f22a8
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
import pytest
......
tests/models/test_registry.py
View file @ cc7f22a8
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
import warnings
......
tests/models/test_transformers.py
View file @ cc7f22a8
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
"""Test the functionality of the Transformers backend."""
from typing import Any, Optional, Union
......
tests/models/test_utils.py
View file @ cc7f22a8
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
import torch
......
tests/models/test_vision.py
View file @ cc7f22a8
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
import pytest
import torch
......
tests/models/utils.py
View file @ cc7f22a8
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
import warnings
from collections.abc import Sequence
from typing import TYPE_CHECKING, Any, NamedTuple, Optional, Union
from typing import Any, NamedTuple, Optional, Union
import torch
import torch.nn.functional as F
......@@ -13,9 +14,6 @@ from vllm.sequence import Logprob, PromptLogprobs, SampleLogprobs
from .registry import HF_EXAMPLE_MODELS
if TYPE_CHECKING:
from ..conftest import HfRunner
TokensText = tuple[list[int], str]
......@@ -317,6 +315,7 @@ def check_embeddings_close(
dim=0)
fail_msg = (f"Test{prompt_idx}:"
f"\nCosine similarity: \t{sim:.4f}"
f"\n{name_0}:\t{embeddings_0[:16]!r}"
f"\n{name_1}:\t{embeddings_1[:16]!r}")
......@@ -337,22 +336,3 @@ class EmbedModelInfo(NamedTuple):
architecture: str = ""
dtype: str = "auto"
enable_test: bool = True
def run_embedding_correctness_test(
hf_model: "HfRunner",
inputs: list[str],
vllm_outputs: Sequence[list[float]],
dimensions: Optional[int] = None,
):
hf_outputs = hf_model.encode(inputs)
if dimensions:
hf_outputs = matryoshka_fy(hf_outputs, dimensions)
check_embeddings_close(
embeddings_0_lst=hf_outputs,
embeddings_1_lst=vllm_outputs,
name_0="hf",
name_1="vllm",
tol=1e-2,
)
tests/mq_llm_engine/conftest.py
View file @ cc7f22a8
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
import pytest
......
tests/mq_llm_engine/test_abort.py
View file @ cc7f22a8
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
"""Test that aborting is handled properly."""
import asyncio
......
tests/mq_llm_engine/test_error_handling.py
View file @ cc7f22a8
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
"""Test that various errors are handled properly."""
import asyncio
......
tests/mq_llm_engine/test_load.py
View file @ cc7f22a8
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
"""Test that the MQLLMEngine is able to handle 10k concurrent requests."""
import asyncio
......
tests/mq_llm_engine/utils.py
View file @ cc7f22a8
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
import asyncio
import multiprocessing
......
tests/multi_step/test_correctness_async_llm.py
View file @ cc7f22a8
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
# Test the AsyncLLMEngine with multi-step-decoding
from typing import Optional
......
tests/multi_step/test_correctness_llm.py
View file @ cc7f22a8
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
# Test the LLMEngine with multi-step-decoding
......
tests/multimodal/test_hasher.py
View file @ cc7f22a8
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
from pathlib import Path
import numpy as np
......
tests/multimodal/test_image.py
View file @ cc7f22a8
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
from pathlib import Path
import numpy as np
......
tests/multimodal/test_inputs.py
View file @ cc7f22a8
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
import torch
......
tests/multimodal/test_processing.py
View file @ cc7f22a8
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
from contextlib import nullcontext
from types import MethodType
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment