Hnimrama/inferencemax uplift restore

cvs/cli_plugins/run_plugin.py

@@ -23,8 +23,12 @@ def get_parser(self, subparsers):
         )
         parser.add_argument(
             "--log-file",
-            default="/tmp/cvs/test.log",
-            help="Pytest: Path to file for logging output (default: /tmp/cvs/test.log)",
+            default=None,
+            metavar="PATH",
+            help=(
+                "Pytest: write logging output to this file (optional). "
+                "Parent directories are created automatically when set."
+            ),
         )
         parser.add_argument(
             "--log-level",

cvs/cli_plugins/unittests/test_run_plugin.py

@@ -79,6 +79,35 @@ def test_run_test_multiple_functions(self, mock_exit, mock_pytest_main):
         mock_pytest_main.assert_called_once_with(expected_args)
         mock_exit.assert_called_once_with(0)
 
+    @patch("cvs.cli_plugins.run_plugin.pytest.main")
+    @patch("cvs.cli_plugins.run_plugin.sys.exit")
+    def test_run_test_omits_log_file_when_not_set(self, mock_exit, mock_pytest_main):
+        """No --log-file is passed to pytest when the user does not request file logging."""
+        args = MagicMock()
+        args.test = "agfhc_cvs"
+        args.function = []
+        args.cluster_file = "/path/to/cluster.json"
+        args.config_file = "/path/to/config.json"
+        args.html = None
+        args.self_contained_html = False
+        args.log_file = None
+        args.log_level = None
+        args.capture = None
+        args.extra_pytest_args = []
+
+        mock_pytest_main.return_value = 0
+
+        with patch.object(self.plugin, "get_test_file", return_value="/mock/path/test.py"):
+            self.plugin.run(args)
+
+        expected_args = [
+            "/mock/path/test.py",
+            "--cluster_file=/path/to/cluster.json",
+            "--config_file=/path/to/config.json",
+        ]
+        mock_pytest_main.assert_called_once_with(expected_args)
+        mock_exit.assert_called_once_with(0)
+
 
 if __name__ == "__main__":
     unittest.main()

cvs/input/config_file/inference/inferencemax_single/mi300x_gpt_oss_120b_single/mi300x_gpt_oss_120b_single_config.json

@@ -0,0 +1,89 @@
+{
+  "_comment": "Single-node InferenceMax (schema_version 1). Set container.image and container.name. ISL+OSL must fit params.max_model_length. Volume bind home only; ContainerOrchestrator adds /home/<user>:/workspace.",
+  "schema_version": 1,
+  "framework": "inferencemax_single",
+  "gpu_arch": "mi300x",
+  "enforce_thresholds": false,
+  "paths": {
+    "shared_fs": "/home/{user-id}",
+    "models_dir": "{shared_fs}/models",
+    "log_dir": "{shared_fs}/LOGS",
+    "hf_token_file": "{shared_fs}/.hf_token"
+  },
+  "model": {
+    "id": "openai/gpt-oss-120b",
+    "remote": 0,
+    "precision": "bf16"
+  },
+  "container": {
+    "lifetime": "per_run",
+    "name": "<changeme>",
+    "image": "<changeme>",
+    "runtime": {
+      "name": "docker",
+      "args": {
+        "network": "host",
+        "ipc": "host",
+        "privileged": true,
+        "shm_size": "128G",
+        "volumes": [
+          "/home/{user-id}:/home/{user-id}"
+        ],
+        "devices": [
+          "/dev/dri",
+          "/dev/kfd"
+        ]
+      }
+    }
+  },
+  "roles": {
+    "server": {
+      "serve_args": {
+        "enforce-eager": true,
+        "gpu-memory-utilization": "0.95",
+        "block-size": 64,
+        "no-enable-prefix-caching": true
+      },
+      "env": {
+        "AMDGCN_USE_BUFFER_OPS": "0",
+        "VLLM_ROCM_USE_AITER": "1",
+        "VLLM_ROCM_QUICK_REDUCE_QUANTIZATION": "INT4"
+      }
+    }
+  },
+  "params": {
+    "backend": "vllm",
+    "base_url": "http://0.0.0.0",
+    "port_no": "8000",
+    "dataset_name": "random",
+    "burstiness": "1.0",
+    "seed": "0",
+    "request_rate": "inf",
+    "random_range_ratio": "0.8",
+    "random_prefix_len": "0",
+    "tensor_parallelism": "8",
+    "tokenizer_mode": "auto",
+    "percentile_metrics": "ttft,tpot,itl,e2el",
+    "metric_percentiles": "99",
+    "num_prompts": "1000",
+    "max_model_length": "8192",
+    "client_poll_count": "50",
+    "client_poll_wait_time": "60",
+    "bench_max_failed_requests": "0"
+  },
+  "sweep": {
+    "sequence_combinations": [
+      {
+        "name": "legacy_profile",
+        "isl": "7168",
+        "osl": "1024"
+      }
+    ],
+    "runs": [
+      {
+        "combo": "legacy_profile",
+        "concurrency": 64
+      }
+    ]
+  }
+}

cvs/input/config_file/inference/inferencemax_single/mi300x_gpt_oss_120b_single/mi300x_gpt_oss_120b_single_threshold.json

@@ -0,0 +1,97 @@
+{
+  "_comment": "client.* thresholds for ISL=7168,OSL=1024,TP=8,CONC=64. Every GATED_METRICS member has a spec so the loader coverage check passes. config.json sets enforce_thresholds=false until calibrated; output_throughput/mean_ttft_ms/mean_tpot_ms carry legacy verify_inference_results targets.",
+  "ISL=7168,OSL=1024,TP=8,CONC=64": {
+    "client.total_token_throughput": {
+      "kind": "min_tok_s",
+      "value": 0
+    },
+    "client.output_throughput": {
+      "kind": "min_tok_s",
+      "value": 4200
+    },
+    "client.mean_ttft_ms": {
+      "kind": "max_ms",
+      "value": 500
+    },
+    "client.median_ttft_ms": {
+      "kind": "max_ms",
+      "value": 1000000
+    },
+    "client.p90_ttft_ms": {
+      "kind": "max_ms",
+      "value": 1000000
+    },
+    "client.p95_ttft_ms": {
+      "kind": "max_ms",
+      "value": 1000000
+    },
+    "client.p99_ttft_ms": {
+      "kind": "max_ms",
+      "value": 1000000
+    },
+    "client.mean_tpot_ms": {
+      "kind": "max_ms",
+      "value": 15
+    },
+    "client.median_tpot_ms": {
+      "kind": "max_ms",
+      "value": 1000000
+    },
+    "client.p90_tpot_ms": {
+      "kind": "max_ms",
+      "value": 1000000
+    },
+    "client.p95_tpot_ms": {
+      "kind": "max_ms",
+      "value": 1000000
+    },
+    "client.p99_tpot_ms": {
+      "kind": "max_ms",
+      "value": 1000000
+    },
+    "client.mean_itl_ms": {
+      "kind": "max_ms",
+      "value": 1000000
+    },
+    "client.median_itl_ms": {
+      "kind": "max_ms",
+      "value": 1000000
+    },
+    "client.p95_itl_ms": {
+      "kind": "max_ms",
+      "value": 1000000
+    },
+    "client.p99_itl_ms": {
+      "kind": "max_ms",
+      "value": 1000000
+    },
+    "client.mean_e2el_ms": {
+      "kind": "max_ms",
+      "value": 1000000
+    },
+    "client.median_e2el_ms": {
+      "kind": "max_ms",
+      "value": 1000000
+    },
+    "client.p90_e2el_ms": {
+      "kind": "max_ms",
+      "value": 1000000
+    },
+    "client.p95_e2el_ms": {
+      "kind": "max_ms",
+      "value": 1000000
+    },
+    "client.p99_e2el_ms": {
+      "kind": "max_ms",
+      "value": 1000000
+    },
+    "client.success_rate": {
+      "kind": "min",
+      "value": 0
+    },
+    "client.failed": {
+      "kind": "max",
+      "value": 1000000000
+    }
+  }
+}