[CI]: reduce HTTP calls inside entrypoints openai tests (#23646)

Signed-off-by: AzizCode92 <azizbenothman76@gmail.com>
Signed-off-by: Aziz <azizbenothman76@gmail.com>
Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>

tests/entrypoints/openai/test_completion.py

@@ -3,8 +3,6 @@
 # imports for guided decoding tests
 import json
 import os
-import shutil
-from tempfile import TemporaryDirectory
 from typing import Optional
 
 import jsonschema
@@ -14,9 +12,7 @@ import pytest_asyncio
 import regex as re
 import requests
 # downloading lora to test lora requests
-from huggingface_hub import snapshot_download
 from openai import BadRequestError
-from transformers import AutoTokenizer
 
 from vllm.transformers_utils.tokenizer import get_tokenizer
 
@@ -26,32 +22,10 @@ from ...utils import RemoteOpenAIServer
 MODEL_NAME = "HuggingFaceH4/zephyr-7b-beta"
 # technically these adapters use a different base model,
 # but we're not testing generation quality here
-LORA_NAME = "typeof/zephyr-7b-beta-lora"
 
 GUIDED_DECODING_BACKENDS = ["outlines", "xgrammar", "guidance"]
 
 
-@pytest.fixture(scope="module")
-def zephyr_lora_files():
-    return snapshot_download(repo_id=LORA_NAME)
-
-
-@pytest.fixture(scope="module")
-def zephyr_lora_added_tokens_files(zephyr_lora_files):
-    tmp_dir = TemporaryDirectory()
-    tmp_model_dir = f"{tmp_dir.name}/zephyr"
-    shutil.copytree(zephyr_lora_files, tmp_model_dir)
-    tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
-    # Copy tokenizer to adapter and add some unique tokens
-    # 32000, 32001, 32002
-    added = tokenizer.add_tokens(["vllm1", "vllm2", "vllm3"],
-                                 special_tokens=True)
-    assert added == 3
-    tokenizer.save_pretrained(tmp_model_dir)
-    yield tmp_model_dir
-    tmp_dir.cleanup()
-
-
 @pytest.fixture(scope="module")
 def default_server_args(zephyr_lora_files, zephyr_lora_added_tokens_files):
     return [