[None][chore] Relocate rlhf_utils.py (#8938)

Signed-off-by: shuyix <219646547+shuyixiong@users.noreply.github.com>

Author: shuyixiong

examples/llm-api/rlhf/rlhf_utils.py renamed to tensorrt_llm/llmapi/rlhf_utils.py

@@ -1,9 +1,7 @@
 import os
 import sys
-from pathlib import Path
 
 import pytest
-from utils.cpp_paths import llm_root  # noqa: F401
 
 from tensorrt_llm._utils import mpi_disabled
 
@@ -23,13 +21,3 @@ def pytest_configure(config):
     pytest.skip(
         "Ray tests are only tested in Ray CI stage or with --run-ray flag",
         allow_module_level=True)
-
-
-@pytest.fixture(scope="function")
-def add_worker_extension_path(llm_root: Path):
-    worker_extension_path = str(llm_root / "examples" / "llm-api" / "rlhf")
-    original_python_path = os.environ.get('PYTHONPATH', '')
-    os.environ['PYTHONPATH'] = os.pathsep.join(
-        filter(None, [worker_extension_path, original_python_path]))
-    yield
-    os.environ['PYTHONPATH'] = original_python_path

tests/unittest/_torch/ray_orchestrator/conftest.py

@@ -139,15 +139,15 @@ def run_generate(llm, hf_model, prompts, sampling_params):
     return llm_logits, ref_logits
 
 
-def test_llm_update_weights(add_worker_extension_path):
+def test_llm_update_weights():
     llama_model_path = str(llm_models_root() / "llama-models-v2/TinyLlama-1.1B-Chat-v1.0")
     kv_cache_config = KvCacheConfig(enable_block_reuse=True, free_gpu_memory_fraction=0.1)
 
     hf_model = HFModel(llama_model_path)
 
     llm = LLM(
         model=llama_model_path,
-        ray_worker_extension_cls="rlhf_utils.WorkerExtension",
+        ray_worker_extension_cls="tensorrt_llm.llmapi.rlhf_utils.WorkerExtension",
         tensor_parallel_size=1,
         pipeline_parallel_size=1,
         kv_cache_config=kv_cache_config,