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": 75575.99123217452,
-      "required_pages_per_second_upper_estimate": 0.006550735374406305
+      "required_token_positions_per_second": 75653.41075888592,
+      "required_pages_per_second_upper_estimate": 0.006557445902763652
     },
     {
       "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": 9446.998904021815,
-      "required_pages_per_second_upper_estimate": 0.0037685491080348714
+      "required_token_positions_per_second": 9456.67634486074,
+      "required_pages_per_second_upper_estimate": 0.0037724095838761527
     },
     {
       "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": 18893.99780804363,
-      "required_pages_per_second_upper_estimate": 0.004370793869886543
+      "required_token_positions_per_second": 18913.35268972148,
+      "required_pages_per_second_upper_estimate": 0.0043752712811284305
     },
     {
       "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": 28340.996712065444,
-      "required_pages_per_second_upper_estimate": 0.0017258180060466934
+      "required_token_positions_per_second": 28370.02903458222,
+      "required_pages_per_second_upper_estimate": 0.0017275859221671408
     },
     {
       "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": 28340.996712065444,
-      "required_pages_per_second_upper_estimate": 0.0021817683797949058
+      "required_token_positions_per_second": 28370.02903458222,
+      "required_pages_per_second_upper_estimate": 0.00218400336834886
     },
     {
       "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": 28340.996712065444,
-      "required_pages_per_second_upper_estimate": 0.3519003279494573
+      "required_token_positions_per_second": 28370.02903458222,
+      "required_pages_per_second_upper_estimate": 0.3522608122301827
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09063380691429757,
-    "producer_p95_seconds": 0.2747461920007481,
-    "gpu_update_seconds": 0.9755479061267306,
-    "producer_batches_per_second": 11.033410534609786,
-    "gpu_batches_per_second": 1.0250649852454226,
-    "producer_headroom": 10.763620544475188,
+    "producer_batch_seconds": 0.07875426600003266,
+    "producer_p95_seconds": 0.29636224299974856,
+    "gpu_update_seconds": 0.9745495842213595,
+    "producer_batches_per_second": 12.697724844512999,
+    "gpu_batches_per_second": 1.0261150547808964,
+    "producer_headroom": 12.37456246777737,
     "sustainable": true,
-    "buffer_seconds": 60.483970179857295,
+    "buffer_seconds": 60.42207422172429,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 483862.00765082374,
-    "upload_bytes_per_second": 4725125.335335737,
+    "generation_bytes_per_second": 490155.8800234727,
+    "upload_bytes_per_second": 4710759.968791011,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.82552311080488,
+    "estimated_full_shards_per_hour": 105.17604160812508,
     "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": 1.2369695059969672,
-      "bytes": 7463645,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 0.9742800499989244,
       "bytes": 772407,
@@ -562,6 +556,12 @@
       "bytes": 8388512,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.2330220080002618,
+      "bytes": 6137294,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -246,3 +246,4 @@
 {"step": 64600, "loss": 0.6922388523817062, "lr": 0.0003, "grad_norm": 0.1990004472183067, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9454299860, "tokens_per_second": 148879.53695223405, "data_wait_s": 8.094000804703683e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 64700, "loss": 0.6381540410220623, "lr": 0.0003, "grad_norm": 0.15739835779292288, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9469111092, "tokens_per_second": 151236.28346579967, "data_wait_s": 8.705999789526686e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 64800, "loss": 0.6068667303770781, "lr": 0.0003, "grad_norm": 0.12981488414858786, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9483709383, "tokens_per_second": 148194.77878463193, "data_wait_s": 8.858998626237735e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 64900, "loss": 0.6011261399835348, "lr": 0.0003, "grad_norm": 0.13736492854896148, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9498208077, "tokens_per_second": 148070.96434491608, "data_wait_s": 1.1184001778019592e-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:d23421e4b19073bf78c3bf4e5f28126e7286c1434b6b9fe025cc5835ccef8308
-size 528561
+oid sha256:258ee811d776c0d7daf73af8fa2be59cc6dbefce51d0a19e886f47901576ecfc
+size 530679