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": 75996.28741390596,
-      "required_pages_per_second_upper_estimate": 0.006587165582207851
+      "required_token_positions_per_second": 75410.86826465954,
+      "required_pages_per_second_upper_estimate": 0.006536422934082441
     },
     {
       "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": 9499.535926738245,
-      "required_pages_per_second_upper_estimate": 0.0037895069118949434
+      "required_token_positions_per_second": 9426.358533082443,
+      "required_pages_per_second_upper_estimate": 0.0037603153554661094
     },
     {
       "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": 18999.07185347649,
-      "required_pages_per_second_upper_estimate": 0.004395100901057394
+      "required_token_positions_per_second": 18852.717066164885,
+      "required_pages_per_second_upper_estimate": 0.004361244296769139
     },
     {
       "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": 28498.607780214737,
-      "required_pages_per_second_upper_estimate": 0.0017354156931756186
+      "required_token_positions_per_second": 28279.075599247324,
+      "required_pages_per_second_upper_estimate": 0.0017220473351510412
     },
     {
       "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": 28498.607780214737,
-      "required_pages_per_second_upper_estimate": 0.002193901716116399
+      "required_token_positions_per_second": 28279.075599247324,
+      "required_pages_per_second_upper_estimate": 0.002177001521121556
     },
     {
       "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": 28498.607780214737,
-      "required_pages_per_second_upper_estimate": 0.3538573299255589
+      "required_token_positions_per_second": 28279.075599247324,
+      "required_pages_per_second_upper_estimate": 0.3511314749648897
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.10981633217207332,
-    "producer_p95_seconds": 0.2631273750012042,
-    "gpu_update_seconds": 0.9701526549375765,
-    "producer_batches_per_second": 9.106113637387569,
-    "gpu_batches_per_second": 1.0307656170505908,
-    "producer_headroom": 8.834320321474822,
+    "producer_batch_seconds": 0.3427846018670948,
+    "producer_p95_seconds": 0.24570008299997426,
+    "gpu_update_seconds": 0.977684008905011,
+    "producer_batches_per_second": 2.917283899431755,
+    "gpu_batches_per_second": 1.0228253616625913,
+    "producer_headroom": 2.852181817910481,
     "sustainable": true,
-    "buffer_seconds": 60.14946460612975,
+    "buffer_seconds": 60.61640855211068,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 389753.6772709055,
+    "generation_bytes_per_second": 361349.18837798125,
     "upload_bytes_per_second": 3964306.7875658553,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.63206614108442,
+    "estimated_full_shards_per_hour": 77.53712404732302,
     "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": [

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -123,3 +123,4 @@
 {"step": 292500, "loss": 0.6153364889323711, "lr": 0.0003, "grad_norm": 0.14834159167227542, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42813718782, "tokens_per_second": 150743.03906782894, "data_wait_s": 1.2544998753583059e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 292600, "loss": 0.6134388819336891, "lr": 0.0003, "grad_norm": 0.1450187929408105, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42828325515, "tokens_per_second": 149952.23042098805, "data_wait_s": 1.2145999789936468e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 292700, "loss": 0.5549955554306507, "lr": 0.0003, "grad_norm": 0.16090119228889782, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42842897664, "tokens_per_second": 149879.54877292723, "data_wait_s": 1.4920999092282727e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 292800, "loss": 0.5693431496620178, "lr": 0.0003, "grad_norm": 0.12763771419192357, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42857511652, "tokens_per_second": 149662.36793654622, "data_wait_s": 7.751001248834655e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:19ab6511360587e31d582d1300eefba2cd0e1f596cabd107f0db36886ce1a1d4
-size 266481
+oid sha256:883d05595cefc6ba454d811b71db4bc84aa95160537484aa4509ff2466c41181
+size 268599