Initial AITuner study orchestrator

configs/examples/capability.example.json

@@ -0,0 +1,14 @@
+{
+  "prefill_service_by_bucket": {
+    "4k": {
+      "tp4_ms": 320,
+      "tp8_ms": 240
+    }
+  },
+  "queueing_knee_by_bucket": {
+    "4k": {
+      "tp4_tok_s_per_gpu": 1000,
+      "tp8_tok_s_per_gpu": 1100
+    }
+  }
+}

configs/examples/study.example.json

@@ -0,0 +1,96 @@
+{
+  "study_id": "example-chat-window",
+  "hardware": {
+    "gpu_count": 8,
+    "gpu_model": "H20",
+    "host_candidates": ["dash0", "dash1"]
+  },
+  "model": {
+    "model_id": "qwen3-30b",
+    "served_model_name": "Qwen/Qwen3-30B-A3B-Instruct-2507"
+  },
+  "engine": {
+    "engine_name": "vllm",
+    "engine_version": "0.x",
+    "exec_path": "/usr/local/bin/vllm",
+    "cwd": ".",
+    "host": "127.0.0.1",
+    "port": 8000,
+    "healthcheck_path": "/v1/models",
+    "ready_timeout_s": 600,
+    "request_timeout_s": 600,
+    "launch_args": [
+      "serve",
+      "/path/to/model"
+    ],
+    "base_envs": {},
+    "base_flags": {
+      "host": "127.0.0.1",
+      "port": 8000,
+      "served-model-name": "Qwen/Qwen3-30B-A3B-Instruct-2507"
+    },
+    "tunable_envs": [
+      "VLLM_ATTENTION_BACKEND",
+      "CUDA_GRAPH_MAX_BATCH_SIZE"
+    ],
+    "tunable_flags": [
+      "tensor-parallel-size",
+      "data-parallel-size",
+      "pipeline-parallel-size",
+      "max-num-seqs",
+      "max-num-batched-tokens",
+      "gpu-memory-utilization",
+      "enable-prefix-caching",
+      "block-size"
+    ],
+    "python_executable": "python3"
+  },
+  "trace": {
+    "windows_path": "trace_windows/windows.json",
+    "window_id": "chat_w_example_peak_0001",
+    "u_field": "sampling_u",
+    "timestamp_field": "timestamp",
+    "max_concurrency": 64
+  },
+  "slo": {
+    "target_pass_rate": 0.95,
+    "ttft_rule": {
+      "kind": "step_ms",
+      "buckets": [
+        {
+          "max_input_tokens": 4096,
+          "threshold_ms": 2000
+        },
+        {
+          "max_input_tokens": 16384,
+          "threshold_ms": 4000
+        },
+        {
+          "threshold_ms": 8000
+        }
+      ]
+    },
+    "tpot_rule": {
+      "kind": "fixed_ms",
+      "threshold_ms": 120
+    }
+  },
+  "search": {
+    "low": 0.0,
+    "high": 1.0,
+    "tolerance": 0.01,
+    "max_probes": 8,
+    "sample_seed": 20260325
+  },
+  "llm": {
+    "system_prompt": "Propose a single engine config patch that increases the maximum feasible sampling_u under the SLO target.",
+    "max_history_trials": 8,
+    "endpoint": {
+      "base_url": "https://example-openai-compatible-endpoint",
+      "model": "gpt-4.1-mini",
+      "api_key_env": "OPENAI_API_KEY",
+      "timeout_s": 120
+    }
+  },
+  "capability_profile_path": "capability.example.json"
+}

configs/examples/trace_windows/traces/chat_w_example_peak_0001.jsonl

@@ -0,0 +1,3 @@
+{"request_id":"example-1","timestamp":0.0,"sampling_u":0.10,"messages":[{"role":"user","content":"hello"}],"input_length":512,"output_length":16}
+{"request_id":"example-2","timestamp":1.0,"sampling_u":0.50,"messages":[{"role":"user","content":"summarize this file"}],"input_length":2048,"output_length":64}
+{"request_id":"example-3","timestamp":2.5,"sampling_u":0.90,"messages":[{"role":"user","content":"write a longer answer"}],"input_length":8192,"output_length":128}

configs/examples/trace_windows/windows.json

@@ -0,0 +1,15 @@
+{
+  "sample_seed": 20260325,
+  "u_field": "sampling_u",
+  "window_duration_seconds": 10.0,
+  "windows": [
+    {
+      "window_id": "chat_w_example_peak_0001",
+      "trace_type": "chat",
+      "trace_file": "traces/chat_w_example_peak_0001.jsonl",
+      "window_start": 0.0,
+      "window_end": 10.0,
+      "num_requests": 3
+    }
+  ]
+}