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": 75551.01885780718,
-      "required_pages_per_second_upper_estimate": 0.006548570832288048
+      "required_token_positions_per_second": 75681.58633512296,
+      "required_pages_per_second_upper_estimate": 0.006559888090301534
     },
     {
       "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": 9443.877357225898,
-      "required_pages_per_second_upper_estimate": 0.0037673038763466962
+      "required_token_positions_per_second": 9460.19829189037,
+      "required_pages_per_second_upper_estimate": 0.003773814541204073
     },
     {
       "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": 18887.754714451796,
-      "required_pages_per_second_upper_estimate": 0.004369349640058767
+      "required_token_positions_per_second": 18920.39658378074,
+      "required_pages_per_second_upper_estimate": 0.004376900762050739
     },
     {
       "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": 28331.632071677694,
-      "required_pages_per_second_upper_estimate": 0.0017252477485795504
+      "required_token_positions_per_second": 28380.59487567111,
+      "required_pages_per_second_upper_estimate": 0.0017282293264547703
     },
     {
       "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": 28331.632071677694,
-      "required_pages_per_second_upper_estimate": 0.002181047463854873
+      "required_token_positions_per_second": 28380.59487567111,
+      "required_pages_per_second_upper_estimate": 0.0021848167560440026
     },
     {
       "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": 28331.632071677694,
-      "required_pages_per_second_upper_estimate": 0.3517840504572767
+      "required_token_positions_per_second": 28380.59487567111,
+      "required_pages_per_second_upper_estimate": 0.3523920046149113
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,39 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09136518590602805,
-    "producer_p95_seconds": 0.29452696799853584,
-    "gpu_update_seconds": 0.9758703603820585,
-    "producer_batches_per_second": 10.945088001337089,
-    "gpu_batches_per_second": 1.0247262757406574,
-    "producer_headroom": 10.68098697227817,
+    "producer_batch_seconds": 0.1801732362186499,
+    "producer_p95_seconds": 0.33215648899931693,
+    "gpu_update_seconds": 0.9741867681463182,
+    "producer_batches_per_second": 5.5502138996185115,
+    "gpu_batches_per_second": 1.0264972104915766,
+    "producer_headroom": 5.406944941390131,
     "sustainable": true,
-    "buffer_seconds": 60.50396234368763,
+    "buffer_seconds": 60.39957962507173,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 384384.7746639473,
-    "upload_bytes_per_second": 6418706.338644549,
+    "generation_bytes_per_second": 411819.44073644676,
+    "upload_bytes_per_second": 6441468.260470278,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.48002462328733,
+    "estimated_full_shards_per_hour": 88.36686531610539,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.1064888179989794,
-      "bytes": 5541578,
-      "pages": 33,
-      "shards": 2
-    },
-    {
-      "seconds": 0.9643621880004503,
-      "bytes": 7410728,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 0.8883736560001125,
       "bytes": 6293157,
@@ -562,6 +550,18 @@
       "bytes": 5985432,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 0.7830291239988583,
+      "bytes": 1207474,
+      "pages": 35,
+      "shards": 1
+    },
+    {
+      "seconds": 1.2103096939972602,
+      "bytes": 12749523,
+      "pages": 22,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -198,3 +198,4 @@
 {"step": 59800, "loss": 0.6514598056674004, "lr": 0.0003, "grad_norm": 0.13604662731343992, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8752313117, "tokens_per_second": 150564.75548546857, "data_wait_s": 1.812400296330452e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 59900, "loss": 0.6550644431263208, "lr": 0.0003, "grad_norm": 0.12457709107079436, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8766957314, "tokens_per_second": 149703.35222456805, "data_wait_s": 1.0060997738037258e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 60000, "loss": 0.6313408296555281, "lr": 0.0003, "grad_norm": 0.13860069394475039, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8781684944, "tokens_per_second": 150619.37458179664, "data_wait_s": 1.96129985852167e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 60100, "loss": 0.5657984018325806, "lr": 0.0003, "grad_norm": 0.13197050722056766, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8796266252, "tokens_per_second": 121109.6552948848, "data_wait_s": 1.123200127040036e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6b8b64864dca4fbed678ad942ab6824e54a06494ab7be284652649d47fa37e8a
-size 426897
+oid sha256:ee644fd9d47fb3225a689e3b22c295431c84c0e2c6eebe28b3d2877c3ccff871
+size 429015