[CI][RL] Consolidate RL entrypoint tests under tests/entrypoints/rl (#59948)

Signed-off-by: wang.yuqi <yuqi.wang@daocloud.io>
Signed-off-by: aoshen02 <aoshen@inferact.ai>
Co-authored-by: wang.yuqi <yuqi.wang@daocloud.io>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
aoshen02
2026-10-07 08:29:09 +00:00
committed by GitHub
co-authored by wang.yuqi Claude Opus 5.5
parent bc8dd4ad27
commit 42ecc8eebe
20 changed files with 46 additions and 22 deletions
@@ -7,7 +7,7 @@ import numpy as np
import pybase64 as base64
import pytest
from ...utils import RemoteOpenAIServer
from tests.utils import RemoteOpenAIServer
MODEL_NAME = "TitanML/tiny-mixtral"
@@ -10,7 +10,7 @@ from unittest.mock import patch
import pytest
import requests
from tests.entrypoints.serve.dev.rlhf.conftest import (
from tests.entrypoints.rl.conftest import (
cached_tokens,
completion_with_cache_details,
gen,
@@ -9,7 +9,7 @@ import requests
import torch
from transformers import AutoModelForCausalLM
from tests.entrypoints.serve.dev.rlhf.conftest import (
from tests.entrypoints.rl.conftest import (
MODEL_NAME,
gen,
health,
@@ -7,7 +7,7 @@ import torch
from prometheus_client.parser import text_string_to_metric_families
from transformers import AutoModelForCausalLM
from tests.entrypoints.serve.dev.rlhf.conftest import MODEL_NAME
from tests.entrypoints.rl.conftest import MODEL_NAME
from tests.utils import RemoteOpenAIServer
from vllm.distributed.weight_transfer import (
HTTPVLLMWeightSyncClient,
@@ -15,6 +15,7 @@ from unittest.mock import patch
import pytest
import torch
from tests.utils import create_new_process_for_each_test
from vllm.config import WeightTransferConfig
from vllm.distributed.weight_transfer.base import (
WeightTransferEngine,
@@ -24,8 +25,6 @@ from vllm.distributed.weight_transfer.base import (
WeightTransferUpdateRequest,
)
from ...utils import create_new_process_for_each_test
# Use a tiny model for fast testing
MODEL_NAME = "hmellor/tiny-random-LlamaForCausalLM"
@@ -5,8 +5,8 @@
Mirrors test_serving_tokens.py but exercises the multimodal piping
using Qwen/Qwen3-VL-2B-Instruct end-to-end via the server's /render ->
/generate -> /detokenize path. Intentionally avoids running the HF
processor in the pytest parent process to keep os.fork() in sibling
tests (e.g. test_weight_transfer_llm.py) deadlock-free.
processor in the pytest parent process to keep os.fork() in other
tests (e.g. entrypoints/rl/weight_transfer/test_offline.py) deadlock-free.
"""
import httpx