Initial commit

results/h200/inference_NVIDIA_H200_flash_attention_3_hopper.json

@@ -0,0 +1,37 @@
+{
+  "model_name": "Qwen/Qwen3-4B",
+  "gpu_name": "NVIDIA H200",
+  "attention_implementation": "flash_attention_3_hopper",
+  "num_requests": 10,
+  "prompt_length": 512,
+  "generation_length": 100,
+  "prefill": {
+    "stage_name": "prefill",
+    "duration_ms": 323.8773119999223,
+    "tokens_processed": 5120,
+    "tokens_per_second": 15808.455270868828,
+    "energy_joules": 98.1449999999968,
+    "energy_per_token": 0.019168945312499373,
+    "avg_power_watts": 250.96736239598317,
+    "peak_memory_gb": 46.1302490234375,
+    "avg_gpu_util_percent": 32.2
+  },
+  "decode": {
+    "stage_name": "decode",
+    "duration_ms": 30558.618001000013,
+    "tokens_processed": 1000,
+    "tokens_per_second": 32.72399294913388,
+    "energy_joules": 4828.459999999999,
+    "energy_per_token": 4.828459999999999,
+    "avg_power_watts": 157.61927190444868,
+    "peak_memory_gb": 46.1302490234375,
+    "avg_gpu_util_percent": 32.2
+  },
+  "e2e_latency_ms": 3088.2495312999936,
+  "e2e_tokens_per_second": 198.17051497855476,
+  "e2e_energy_joules": 4926.604999999996,
+  "e2e_energy_per_token": 0.8050008169934634,
+  "ttft_ms": 32.38773119999223,
+  "itl_ms": 30.558618001000013,
+  "timestamp": 1768541964.4743361
+}

results/h200/pretrain_NVIDIA_H200_flash_attention_3_hopper.json

@@ -0,0 +1,47 @@
+{
+  "model_name": "Qwen/Qwen3-4B",
+  "gpu_name": "NVIDIA H200",
+  "attention_implementation": "flash_attention_3_hopper",
+  "batch_size": 3,
+  "sequence_length": 2048,
+  "num_steps": 10,
+  "forward": {
+    "stage_name": "forward",
+    "duration_ms": 1605.9521619997668,
+    "tokens_processed": 61440,
+    "tokens_per_second": 38257.67756587068,
+    "energy_joules": 817.7539999999863,
+    "energy_per_token": 0.01330979817708311,
+    "avg_power_watts": 476.6091506406698,
+    "peak_memory_gb": 76.5540771484375,
+    "avg_gpu_util_percent": 95.1
+  },
+  "backward": {
+    "stage_name": "backward",
+    "duration_ms": 3448.8081949999696,
+    "tokens_processed": 61440,
+    "tokens_per_second": 17814.849804948502,
+    "energy_joules": 1765.182000000008,
+    "energy_per_token": 0.02873017578125013,
+    "avg_power_watts": 498.84691252245983,
+    "peak_memory_gb": 76.5540771484375,
+    "avg_gpu_util_percent": 95.1
+  },
+  "optimizer": {
+    "stage_name": "optimizer",
+    "duration_ms": 545.701982000196,
+    "tokens_processed": 61440,
+    "tokens_per_second": 112588.92587268984,
+    "energy_joules": 332.4770000000135,
+    "energy_per_token": 0.005411409505208553,
+    "avg_power_watts": 521.4900438388863,
+    "peak_memory_gb": 76.5540771484375,
+    "avg_gpu_util_percent": 95.1
+  },
+  "total_duration_ms": 5600.462338999932,
+  "total_tokens": 61440,
+  "total_tokens_per_second": 10970.522839186035,
+  "total_energy_joules": 2915.4130000000077,
+  "total_energy_per_token": 0.047451383463541795,
+  "timestamp": 1768541921.6000674
+}