[ROCm][CI] Cleaning and restructuring amd-ci legacy pipeline (#34839)

Signed-off-by: Andreas Karatzas <akaratza@amd.com>

tests/evals/gpt_oss/configs/models-gfx942.txt

@@ -1,3 +1,3 @@
 # GFX942 model configurations for GPQA evaluation
 # Tests different environment variable combinations
-gpt-oss-20b-rocm-baseline.yaml
+gpt-oss-20b-rocm-baseline.yaml

tests/evals/gsm8k/configs/DeepSeek-R1-DP_MI325.yaml

@@ -0,0 +1,12 @@
+model_name: "deepseek-ai/DeepSeek-R1"
+accuracy_threshold: 0.95
+num_questions: 1319
+num_fewshot: 5
+startup_max_wait_seconds: 1200
+server_args: >-
+  --enforce-eager
+  --max-model-len 4096
+  --data-parallel-size 8
+  --enable-expert-parallel
+  --attention-backend=TRITON_ATTN
+  --speculative-config '{"method":"mtp","num_speculative_tokens":3}'

tests/evals/gsm8k/configs/DeepSeek-R1-TP_MI325.yaml

@@ -0,0 +1,12 @@
+model_name: "deepseek-ai/DeepSeek-R1"
+accuracy_threshold: 0.95
+num_questions: 1319
+num_fewshot: 5
+startup_max_wait_seconds: 1200
+server_args: >-
+  --enforce-eager
+  --max-model-len 4096
+  --tensor-parallel-size 8
+  --enable-expert-parallel
+  --attention-backend=TRITON_ATTN
+  --speculative-config '{"method":"mtp","num_speculative_tokens":3}'

tests/evals/gsm8k/configs/DeepSeek-V3.2-DP_MI325.yaml

@@ -0,0 +1,12 @@
+model_name: "deepseek-ai/DeepSeek-V3.2"
+accuracy_threshold: 0.95
+num_questions: 1319
+num_fewshot: 5
+startup_max_wait_seconds: 1200
+server_args: >-
+  --enforce-eager
+  --max-model-len 4096
+  --data-parallel-size 8
+  --enable-expert-parallel
+  --attention-backend=TRITON_ATTN
+  --speculative-config '{"method":"mtp","num_speculative_tokens":3}'

tests/evals/gsm8k/configs/DeepSeek-V3.2-TP_MI325.yaml

@@ -0,0 +1,12 @@
+model_name: "deepseek-ai/DeepSeek-V3.2"
+accuracy_threshold: 0.95
+num_questions: 1319
+num_fewshot: 5
+startup_max_wait_seconds: 1200
+server_args: >-
+  --enforce-eager
+  --max-model-len 4096
+  --tensor-parallel-size 8
+  --enable-expert-parallel
+  --attention-backend=TRITON_ATTN
+  --speculative-config '{"method":"mtp","num_speculative_tokens":3}'

tests/evals/gsm8k/configs/models-mi3xx.txt

@@ -0,0 +1,4 @@
+DeepSeek-R1-TP_MI325.yaml
+DeepSeek-R1-DP_MI325.yaml
+DeepSeek-V3.2-TP_MI325.yaml
+DeepSeek-V3.2-DP_MI325.yaml

tests/evals/gsm8k/test_gsm8k_correctness.py

@@ -64,6 +64,16 @@ def test_gsm8k_correctness(config_filename):
             "Marlin kernels are not supported."
         )
 
+    # TODO(akaratza): Enable DeepSeek-V3.2 and DeepSeek-R1 on ROCm platforms
+    if current_platform.is_rocm() and (
+        "deepseek-ai/DeepSeek-V3.2" in eval_config["model_name"]
+        or "deepseek-ai/DeepSeek-R1" in eval_config["model_name"]
+    ):
+        pytest.skip(
+            "Skipping DeepSeek-V3.2 and DeepSeek-R1 on ROCm platforms "
+            "due to agent pool disk space issues and pod evictions."
+        )
+
     # Parse server arguments from config (use shlex to handle quoted strings)
     server_args_str = eval_config.get("server_args", "")
     server_args = shlex.split(server_args_str) if server_args_str else []

tests/quantization/test_mi3xx_moe.py

@@ -0,0 +1,6 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+
+def test_mi3xx_moe():
+    print("TODO: add tests for Mi3xx MoE quantization")

tests/rocm/aiter/test_mla_fp8_support_check.py

@@ -31,7 +31,7 @@ class TestAiterMlaFp8SupportCheck:
 
         # Should return False without raising
         with patch(
-            "vllm._aiter_ops.inspect.signature",
+            "inspect.signature",
             side_effect=ImportError("No module"),
         ):
             result = _check_aiter_mla_fp8_support()
@@ -46,7 +46,7 @@ class TestAiterMlaFp8SupportCheck:
         aiter_ops._AITER_MLA_SUPPORTS_FP8 = None
 
         with patch(
-            "vllm._aiter_ops.inspect.signature",
+            "inspect.signature",
             side_effect=ModuleNotFoundError("Module not found"),
         ):
             # Should return False without raising
@@ -63,7 +63,7 @@ class TestAiterMlaFp8SupportCheck:
         aiter_ops._AITER_MLA_SUPPORTS_FP8 = None
 
         with patch(
-            "vllm._aiter_ops.inspect.signature",
+            "inspect.signature",
             side_effect=AttributeError("No attribute"),
         ):
             assert _check_aiter_mla_fp8_support() is False
@@ -78,7 +78,7 @@ class TestAiterMlaFp8SupportCheck:
         aiter_ops._AITER_MLA_SUPPORTS_FP8 = None
 
         with patch(
-            "vllm._aiter_ops.inspect.signature",
+            "inspect.signature",
             side_effect=ValueError("No signature"),
         ):
             assert _check_aiter_mla_fp8_support() is False
@@ -93,7 +93,7 @@ class TestAiterMlaFp8SupportCheck:
         aiter_ops._AITER_MLA_SUPPORTS_FP8 = None
 
         with patch(
-            "vllm._aiter_ops.inspect.signature",
+            "inspect.signature",
             side_effect=TypeError("Not a callable"),
         ):
             assert _check_aiter_mla_fp8_support() is False