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": 75079.50577072128,
-      "required_pages_per_second_upper_estimate": 0.006507701272938436
+      "required_token_positions_per_second": 75231.6824417679,
+      "required_pages_per_second_upper_estimate": 0.0065208915610965095
     },
     {
       "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": 9384.93822134016,
-      "required_pages_per_second_upper_estimate": 0.003743792173823265
+      "required_token_positions_per_second": 9403.960305220988,
+      "required_pages_per_second_upper_estimate": 0.0037513803674888253
     },
     {
       "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": 18769.87644268032,
-      "required_pages_per_second_upper_estimate": 0.004342080576471163
+      "required_token_positions_per_second": 18807.920610441975,
+      "required_pages_per_second_upper_estimate": 0.004350881425128338
     },
     {
       "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": 28154.814664020483,
-      "required_pages_per_second_upper_estimate": 0.0017144804961424687
+      "required_token_positions_per_second": 28211.880915662965,
+      "required_pages_per_second_upper_estimate": 0.0017179555314640125
     },
     {
       "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": 28154.814664020483,
-      "required_pages_per_second_upper_estimate": 0.0021674355703515096
+      "required_token_positions_per_second": 28211.880915662965,
+      "required_pages_per_second_upper_estimate": 0.002171828688372443
     },
     {
       "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": 28154.814664020483,
-      "required_pages_per_second_upper_estimate": 0.34958857002397015
+      "required_token_positions_per_second": 28211.880915662965,
+      "required_pages_per_second_upper_estimate": 0.35029714188091143
     }
   ],
   "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": 0.45847250633593717,
-    "producer_p95_seconds": 0.20123779899950023,
-    "gpu_update_seconds": 0.9819990054962731,
-    "producer_batches_per_second": 2.18115587342825,
-    "gpu_batches_per_second": 1.0183309701975,
-    "producer_headroom": 2.1418928985388965,
+    "producer_batch_seconds": 0.08266829907030626,
+    "producer_p95_seconds": 0.19528255600016564,
+    "gpu_update_seconds": 0.9800126437032455,
+    "producer_batches_per_second": 12.096535325464213,
+    "gpu_batches_per_second": 1.0203949983963747,
+    "producer_headroom": 11.854757563957882,
     "sustainable": true,
-    "buffer_seconds": 60.88393834076893,
+    "buffer_seconds": 60.76078390960122,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 297517.4297877964,
-    "upload_bytes_per_second": 3673096.1446132744,
+    "generation_bytes_per_second": 295724.95407618047,
+    "upload_bytes_per_second": 3709914.971222165,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.840314581159774,
+    "estimated_full_shards_per_hour": 63.45569102014599,
     "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": 1.0497820410000713,
-      "bytes": 5934890,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.250543007000033,
       "bytes": 8195692,
@@ -562,6 +556,12 @@
       "bytes": 11603941,
       "pages": 24,
       "shards": 4
+    },
+    {
+      "seconds": 1.57969898599913,
+      "bytes": 12094782,
+      "pages": 45,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -97,3 +97,4 @@
 {"step": 246800, "loss": 0.6446951851248741, "lr": 0.0003, "grad_norm": 0.14357716821001443, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36124434723, "tokens_per_second": 149976.7950049897, "data_wait_s": 1.3602000763057731e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 246900, "loss": 0.5773094110190868, "lr": 0.0003, "grad_norm": 0.1615344566924537, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36139181982, "tokens_per_second": 151018.7287214966, "data_wait_s": 1.038899972627405e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 247000, "loss": 0.5580745562911034, "lr": 0.0003, "grad_norm": 0.14490235312005187, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36153740645, "tokens_per_second": 132000.8356372005, "data_wait_s": 1.1910999091924168e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 247100, "loss": 0.5960701331496239, "lr": 0.0003, "grad_norm": 0.15297610330697223, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36168429777, "tokens_per_second": 149928.20502250365, "data_wait_s": 1.576699833094608e-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:d95d264bd934a504659d9700b37dedcb1035159478e36d39cd5f3c3e045efaf0
-size 211413
+oid sha256:afc915e76e6a841523b7c90cc193e74ea0e7f0e5d5d244b577d71d1296eced91
+size 213531