Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75811.29955144729,
-      "required_pages_per_second_upper_estimate": 0.006571131303137385
+      "required_token_positions_per_second": 76008.31219904328,
+      "required_pages_per_second_upper_estimate": 0.0065882078601070156
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9476.41244393091,
-      "required_pages_per_second_upper_estimate": 0.003780282608876221
+      "required_token_positions_per_second": 9501.03902488041,
+      "required_pages_per_second_upper_estimate": 0.0037901065202171737
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18952.82488786182,
-      "required_pages_per_second_upper_estimate": 0.004384402479481238
+      "required_token_positions_per_second": 19002.07804976082,
+      "required_pages_per_second_upper_estimate": 0.004395796331660528
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28429.237331792734,
-      "required_pages_per_second_upper_estimate": 0.0017311913968253332
+      "required_token_positions_per_second": 28503.117074641228,
+      "required_pages_per_second_upper_estimate": 0.0017356902855477523
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28429.237331792734,
-      "required_pages_per_second_upper_estimate": 0.0021885613869671817
+      "required_token_positions_per_second": 28503.117074641228,
+      "required_pages_per_second_upper_estimate": 0.0021942488540838763
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28429.237331792734,
-      "required_pages_per_second_upper_estimate": 0.35299598112411357
+      "required_token_positions_per_second": 28503.117074641228,
+      "required_pages_per_second_upper_estimate": 0.3539133202706983
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.053332340343729356,
-    "producer_p95_seconds": 0.15089832299963746,
-    "gpu_update_seconds": 0.9725199335221326,
-    "producer_batches_per_second": 18.75034910440747,
-    "gpu_batches_per_second": 1.0282565585862533,
-    "producer_headroom": 18.23508826453513,
+    "producer_batch_seconds": 0.05806758685935165,
+    "producer_p95_seconds": 0.1541919940000298,
+    "gpu_update_seconds": 0.9699991733394656,
+    "producer_batches_per_second": 17.22131147660999,
+    "gpu_batches_per_second": 1.0309287136371972,
+    "producer_headroom": 16.704657896133142,
     "sustainable": true,
-    "buffer_seconds": 60.29623587837222,
+    "buffer_seconds": 60.139948747046866,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790696019647648547-3fdea0a8/train.jsonl

@@ -35,3 +35,4 @@
 {"step": 278500, "loss": 0.6061176229268312, "lr": 0.0003, "grad_norm": 0.16668151308511772, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40765085288, "tokens_per_second": 152982.57103350223, "data_wait_s": 1.4629999895987567e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 278600, "loss": 0.5726190768182278, "lr": 0.0003, "grad_norm": 0.13401794960571295, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40779598508, "tokens_per_second": 149480.0919833245, "data_wait_s": 1.6554000467294827e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 278700, "loss": 0.5746679967269301, "lr": 0.0003, "grad_norm": 0.15670635691392462, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40794039350, "tokens_per_second": 148513.58695770724, "data_wait_s": 1.127800078393193e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 278800, "loss": 0.6013021990656853, "lr": 0.0003, "grad_norm": 0.16295544019408925, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40808645024, "tokens_per_second": 150309.74284495253, "data_wait_s": 1.3943999874754809e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790696019.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.22407.0.1790696019647648547-3fdea0a8

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8f5233bb13a9fb9924c7b4b02df7c7f649453045229a7de706fd28e459969d53
-size 78531
+oid sha256:9c1219223534b9f9453fad7b8c2ec0c18976193f48bbf6c0c0071c5a2bfd6078
+size 80649