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": 75518.63492565605,
-      "required_pages_per_second_upper_estimate": 0.006545763875125513
+      "required_token_positions_per_second": 75504.59572267949,
+      "required_pages_per_second_upper_estimate": 0.0065445469925935395
     },
     {
       "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": 9439.829365707006,
-      "required_pages_per_second_upper_estimate": 0.0037656890720069434
+      "required_token_positions_per_second": 9438.074465334936,
+      "required_pages_per_second_upper_estimate": 0.003764989016010426
     },
     {
       "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": 18879.658731414012,
-      "required_pages_per_second_upper_estimate": 0.00436747677686741
+      "required_token_positions_per_second": 18876.14893066987,
+      "required_pages_per_second_upper_estimate": 0.004366664846235637
     },
     {
       "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": 28319.488097121015,
-      "required_pages_per_second_upper_estimate": 0.001724508244243559
+      "required_token_positions_per_second": 28314.223396004807,
+      "required_pages_per_second_upper_estimate": 0.001724187651567334
     },
     {
       "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": 28319.488097121015,
-      "required_pages_per_second_upper_estimate": 0.002180112587076827
+      "required_token_positions_per_second": 28314.223396004807,
+      "required_pages_per_second_upper_estimate": 0.0021797072958112756
     },
     {
       "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": 28319.488097121015,
-      "required_pages_per_second_upper_estimate": 0.35163326293655106
+      "required_token_positions_per_second": 28314.223396004807,
+      "required_pages_per_second_upper_estimate": 0.351567892968509
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.3396746921015108,
-    "producer_p95_seconds": 0.24990330199943855,
-    "gpu_update_seconds": 0.9762888335121679,
-    "producer_batches_per_second": 2.9439932478135664,
-    "gpu_batches_per_second": 1.0242870405498052,
-    "producer_headroom": 2.8741877337756057,
+    "producer_batch_seconds": 0.08245267762484332,
+    "producer_p95_seconds": 0.2489672019983118,
+    "gpu_update_seconds": 0.9764703630861791,
+    "producer_batches_per_second": 12.128168894040817,
+    "gpu_batches_per_second": 1.0240966216726275,
+    "producer_headroom": 11.84279748353454,
     "sustainable": true,
-    "buffer_seconds": 60.52990767775441,
+    "buffer_seconds": 60.5411625113431,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 354276.5828814239,
-    "upload_bytes_per_second": 3955706.1290865983,
+    "generation_bytes_per_second": 370413.8736464335,
+    "upload_bytes_per_second": 3967476.8327816944,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.01950754959142,
+    "estimated_full_shards_per_hour": 79.48219449085956,
     "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": 0.9434353479991842,
-      "bytes": 5975783,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 0.9842061780000222,
       "bytes": 8165489,
@@ -562,6 +556,12 @@
       "bytes": 5726191,
       "pages": 20,
       "shards": 2
+    },
+    {
+      "seconds": 1.1066016949989717,
+      "bytes": 7887477,
+      "pages": 44,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -126,3 +126,4 @@
 {"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}
 {"step": 292900, "loss": 0.5454485975205898, "lr": 0.0003, "grad_norm": 0.15755913143376335, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42872125397, "tokens_per_second": 150112.70885700796, "data_wait_s": 1.3035998563282192e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 293000, "loss": 0.5930706281214952, "lr": 0.0003, "grad_norm": 0.1754426991099345, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42886873471, "tokens_per_second": 151069.63870371747, "data_wait_s": 9.398998372489586e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 293100, "loss": 0.5831968560814857, "lr": 0.0003, "grad_norm": 0.15030196289962997, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42901558737, "tokens_per_second": 150688.91296613295, "data_wait_s": 1.6109999705804512e-05, "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:d36f33b315aa949714fe3ae7539e4c9f265a60ea162724d3034eda6edb24add5
-size 272835
+oid sha256:c32b88e6d908220643a013537a7b323b88a4d7801bbf631a5cb932219f21328f
+size 274953