[test] Switch to TorchSampler in tests

Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>

Author: Funatiq

tests/unittest/_torch/modeling/test_modeling_nemotron_h.py

@@ -56,7 +56,6 @@ def create_nemotron_h_llm(model_folder,
         kv_cache_config=KvCacheConfig(
             mamba_ssm_cache_dtype=mamba_ssm_cache_dtype)
         if mamba_ssm_cache_dtype is not None else KvCacheConfig(),
-        sampler_type="TRTLLMSampler",
         enable_chunked_prefill=enable_chunked_prefill,
         **kwargs,
     )

tests/unittest/_torch/ray_orchestrator/multi_gpu/test_accuracy_with_allreduce_strategy.py

@@ -102,9 +102,8 @@ def compare_logprobs(logprobs_list, ref_new_token_logprobs_list):
 
 @pytest.mark.gpu4
 @pytest.mark.parametrize("model_dir", ["Qwen2-7B-Instruct"])
-@pytest.mark.parametrize("sampler_type", ["TRTLLMSampler"])
 @pytest.mark.parametrize("allreduce_strategy", ["NCCL", "AUTO"])
-def test_accuracy_with_allreduce_strategy(model_dir, sampler_type, allreduce_strategy):
+def test_accuracy_with_allreduce_strategy(model_dir, allreduce_strategy):
     """Test accuracy with different allreduce strategies.
 
     This test validates that both NCCL and AUTO allreduce strategies produce
@@ -138,7 +137,6 @@ def test_accuracy_with_allreduce_strategy(model_dir, sampler_type, allreduce_str
         max_batch_size=256,
         max_num_tokens=8192,
         tensor_parallel_size=4,
-        sampler_type=sampler_type,
         allreduce_strategy=allreduce_strategy,
     )