Upload workloads/synthetic/02_optimized/cached_metrics.json with huggingface_hub
Browse files
workloads/synthetic/02_optimized/cached_metrics.json
ADDED
|
@@ -0,0 +1,29 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"steps": 50,
|
| 3 |
+
"tokens_per_sec": 318.0,
|
| 4 |
+
"mfu_pct": 51.0,
|
| 5 |
+
"hbm_peak_gb": 168.0,
|
| 6 |
+
"hbm_avg_gb": 152.0,
|
| 7 |
+
"gpu_util_pct": 86.0,
|
| 8 |
+
"top_kernels": [
|
| 9 |
+
{"name": "aten::matmul (bf16)", "pct_time": 46.0},
|
| 10 |
+
{"name": "flash_attn_rocm_fwd", "pct_time": 19.0},
|
| 11 |
+
{"name": "flash_attn_rocm_bwd", "pct_time": 14.0},
|
| 12 |
+
{"name": "aten::layer_norm", "pct_time": 7.0},
|
| 13 |
+
{"name": "aten::cross_entropy_loss", "pct_time": 4.0}
|
| 14 |
+
],
|
| 15 |
+
"attention_kernel_loaded": "flash_rocm",
|
| 16 |
+
"waste_budget": {
|
| 17 |
+
"useful_gpu": 0.78,
|
| 18 |
+
"data_wait": 0.04,
|
| 19 |
+
"host_gap": 0.05,
|
| 20 |
+
"comm_excess": 0.0,
|
| 21 |
+
"memory_headroom": 0.04,
|
| 22 |
+
"precision_path": 0.02,
|
| 23 |
+
"kernel_shape": 0.03
|
| 24 |
+
},
|
| 25 |
+
"warnings": [],
|
| 26 |
+
"rocm_version": "6.1.0",
|
| 27 |
+
"pytorch_version": "2.3.0+rocm6.1",
|
| 28 |
+
"runner_kind": "fake"
|
| 29 |
+
}
|