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": 75570.71051583829,
-      "required_pages_per_second_upper_estimate": 0.006550277655298134
+      "required_token_positions_per_second": 75326.08513115185,
+      "required_pages_per_second_upper_estimate": 0.006529074146950881
     },
     {
       "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": 9446.338814479786,
-      "required_pages_per_second_upper_estimate": 0.003768285788447338
+      "required_token_positions_per_second": 9415.760641393981,
+      "required_pages_per_second_upper_estimate": 0.0037560876980189807
     },
     {
       "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": 18892.677628959573,
-      "required_pages_per_second_upper_estimate": 0.004370488469689805
+      "required_token_positions_per_second": 18831.521282787962,
+      "required_pages_per_second_upper_estimate": 0.004356341025317935
     },
     {
       "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": 28339.01644343936,
-      "required_pages_per_second_upper_estimate": 0.0017256974180770313
+      "required_token_positions_per_second": 28247.28192418194,
+      "required_pages_per_second_upper_estimate": 0.0017201112671480809
     },
     {
       "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": 28339.01644343936,
-      "required_pages_per_second_upper_estimate": 0.0021816159332343353
+      "required_token_positions_per_second": 28247.28192418194,
+      "required_pages_per_second_upper_estimate": 0.0021745539560044974
     },
     {
       "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": 28339.01644343936,
-      "required_pages_per_second_upper_estimate": 0.3518757396406541
+      "required_token_positions_per_second": 28247.28192418194,
+      "required_pages_per_second_upper_estimate": 0.35073670392716316
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.129124772499992,
-    "producer_p95_seconds": 0.24966627399953722,
-    "gpu_update_seconds": 0.9756160752855156,
-    "producer_batches_per_second": 7.744447332908656,
-    "gpu_batches_per_second": 1.0249933609461575,
-    "producer_headroom": 7.555607312187723,
+    "producer_batch_seconds": 0.07883505268755187,
+    "producer_p95_seconds": 0.2057646540015412,
+    "gpu_update_seconds": 0.9787844393032057,
+    "producer_batches_per_second": 12.68471277571558,
+    "gpu_batches_per_second": 1.0216754168179232,
+    "producer_headroom": 12.415599481900983,
     "sustainable": true,
-    "buffer_seconds": 60.48819666770197,
+    "buffer_seconds": 60.684635236798755,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 382211.29352878686,
-    "upload_bytes_per_second": 4854486.8576622605,
+    "generation_bytes_per_second": 397134.3323118052,
+    "upload_bytes_per_second": 4800874.913550711,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.01364616773323,
+    "estimated_full_shards_per_hour": 85.21578290000551,
     "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": 0.7338976749997528,
-      "bytes": 1218142,
-      "pages": 34,
-      "shards": 1
-    },
-    {
-      "seconds": 1.1807161469996572,
-      "bytes": 7525394,
-      "pages": 28,
-      "shards": 3
-    },
     {
       "seconds": 1.1362227980007447,
       "bytes": 11027017,
@@ -562,6 +550,18 @@
       "bytes": 7507792,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 2.5084295430006023,
+      "bytes": 11998594,
+      "pages": 38,
+      "shards": 3
+    },
+    {
+      "seconds": 1.094205030000012,
+      "bytes": 275280,
+      "pages": 8,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -127,3 +127,4 @@
 {"step": 249800, "loss": 0.5844248123466969, "lr": 0.0003, "grad_norm": 0.15406217257925406, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36563913568, "tokens_per_second": 150778.13998014262, "data_wait_s": 8.314998922287486e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 249900, "loss": 0.5893625961616635, "lr": 0.0003, "grad_norm": 0.16091751731171505, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36578524362, "tokens_per_second": 149190.4559398415, "data_wait_s": 1.704299938865006e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 250000, "loss": 0.5805718433111906, "lr": 0.0003, "grad_norm": 0.15552392142350027, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36593230586, "tokens_per_second": 150453.01984795262, "data_wait_s": 1.3057999240118079e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 250100, "loss": 0.568646140396595, "lr": 0.0003, "grad_norm": 0.15801772858695023, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36607938466, "tokens_per_second": 136372.44597009715, "data_wait_s": 1.2935999620822258e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:46d4966ddfb6427e974be47f889a8a61b636b34333c4a479300ef4f3ca981739
-size 275736
+oid sha256:5ea0ab08ca35e7575ef9e265501f4a0c4473772d8cca4cfeb8d48823e77b9575
+size 277854