[ROCm][CI] Fix and stabilize EAGLE3 acceptance tests

tests/v1/spec_decode/test_acceptance_length.py

@@ -39,6 +39,10 @@ class Eagle3ModelConfig:
     marks: list = field(default_factory=list)
     # Custom relative tolerance (defaults to DEFAULT_RTOL if None)
     rtol: float | None = None
+    # ROCm-specific test configuration
+    rocm_tp_sizes: set[int] | None = None
+    rocm_expert_parallel_tp_sizes: set[int] = field(default_factory=set)
+    rocm_expected_acceptance_lengths_per_pos: list[float] = field(default_factory=list)
 
 
 # Model configurations for EAGLE3 acceptance length tests.
@@ -69,6 +73,7 @@ class Eagle3ModelConfig:
         # FLASHINFER incompatible: gpt-oss-20b uses sink attention which
         # FLASHINFER does not support ("sink setting not supported")
         excluded_backends={AttentionBackendEnum.FLASHINFER},
+        rocm_expected_acceptance_lengths_per_pos=[0.7040, 0.4820, 0.3350],
     ),
     Eagle3ModelConfig(
         verifier="Qwen/Qwen3-VL-30B-A3B-Instruct-FP8",
@@ -80,6 +85,8 @@ class Eagle3ModelConfig:
             pytest.mark.slow_test,
         ],
         rtol=0.15,  # Higher tolerance due to small absolute values at position 2
+        rocm_tp_sizes={4},
+        rocm_expert_parallel_tp_sizes={4},
     ),
 ]
 
@@ -99,16 +106,14 @@ class Eagle3ModelConfig:
 
 
 def get_available_attention_backends() -> list[str]:
+    if current_platform.is_rocm():
+        return ["auto"]
+
     # Check if get_valid_backends is actually defined in the platform class
     # (not just returning None from __getattr__)
     get_valid_backends = getattr(current_platform.__class__, "get_valid_backends", None)
     if get_valid_backends is None:
-        if current_platform.is_rocm():
-            # ROCm uses Triton as its default attention backend since
-            # Flash Attention is not supported.
-            return ["TRITON_ATTN"]
-        else:
-            return ["FLASH_ATTN"]
+        return ["FLASH_ATTN"]
 
     device_capability = current_platform.get_device_capability()
     if device_capability is None:
@@ -167,6 +172,8 @@ def get_mt_bench_prompts(
         disable_shuffle=False,
         skip_chat_template=False,
         trust_remote_code=False,
+        enable_multimodal_chat=False,
+        request_id_prefix="",
     )
     samples = get_samples(args, tokenizer)
     prompt_ids = [
@@ -233,9 +240,19 @@ def test_eagle3_acceptance_length(
     monkeypatch: pytest.MonkeyPatch,
 ):
     # Skip if this backend is incompatible with the model
-    backend_enum = AttentionBackendEnum[attention_backend]
-    if backend_enum in model_config.excluded_backends:
-        pytest.skip(f"{attention_backend} is incompatible with {model_config.id}")
+    if (
+        current_platform.is_rocm()
+        and model_config.rocm_tp_sizes is not None
+        and tp_size not in model_config.rocm_tp_sizes
+    ):
+        tp_ids = ", ".join(f"tp{tp}" for tp in sorted(model_config.rocm_tp_sizes))
+        pytest.skip(f"{model_config.id} is validated with {tp_ids} on ROCm")
+    attention_config = None
+    if attention_backend != "auto":
+        backend_enum = AttentionBackendEnum[attention_backend]
+        if backend_enum in model_config.excluded_backends:
+            pytest.skip(f"{attention_backend} is incompatible with {model_config.id}")
+        attention_config = {"backend": attention_backend}
 
     with monkeypatch.context() as m:
         m.setenv("VLLM_ALLOW_INSECURE_SERIALIZATION", "1")
@@ -247,11 +264,15 @@ def test_eagle3_acceptance_length(
                 "model": model_config.drafter,
                 "num_speculative_tokens": num_spec_tokens,
             },
-            attention_config={"backend": attention_backend},
+            attention_config=attention_config,
             tensor_parallel_size=tp_size,
             gpu_memory_utilization=0.7,
             disable_log_stats=False,
             max_model_len=DEFAULT_MAX_MODEL_LEN,
+            enable_expert_parallel=(
+                current_platform.is_rocm()
+                and tp_size in model_config.rocm_expert_parallel_tp_sizes
+            ),
         ) as vllm_runner:
             tokenizer = vllm_runner.llm.get_tokenizer()
             prompt_ids = get_mt_bench_prompts(tokenizer, DEFAULT_NUM_PROMPTS)
@@ -272,6 +293,11 @@ def test_eagle3_acceptance_length(
             expected = model_config.expected_acceptance_length
             actual_per_pos = results["acceptance_lengths_per_pos"]
             expected_per_pos = model_config.expected_acceptance_lengths_per_pos
+            if (
+                current_platform.is_rocm()
+                and model_config.rocm_expected_acceptance_lengths_per_pos
+            ):
+                expected_per_pos = model_config.rocm_expected_acceptance_lengths_per_pos
 
             rel_error = abs(actual_acceptance_length - expected) / expected
 
@@ -294,14 +320,14 @@ def test_eagle3_acceptance_length(
                     zip(actual_per_pos, expected_per_pos)
                 ):
                     if exp > 0:
-                        pos_rel_error = abs(actual - exp) / exp
-                        assert pos_rel_error <= rtol, (
+                        min_expected = exp * (1 - rtol)
+                        assert actual >= min_expected, (
                             f"Per-position acceptance length regression at pos {pos} "
                             f"for {model_config.id}!\n"
                             f"  Expected: {exp:.3f}\n"
                             f"  Actual:   {actual:.3f}\n"
-                            f"  Relative error: {pos_rel_error:.2%} "
-                            f"(tolerance: {rtol:.2%})"
+                            f"  Minimum:  {min_expected:.3f}\n"
+                            f"  Tolerance: rtol={rtol:.2%}"
                         )
 
             print(