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": 75828.00041081537,
-      "required_pages_per_second_upper_estimate": 0.006572578891299469
+      "required_token_positions_per_second": 75488.0503807444,
+      "required_pages_per_second_upper_estimate": 0.006543112884288392
     },
     {
       "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": 9478.50005135192,
-      "required_pages_per_second_upper_estimate": 0.0037811153866889745
+      "required_token_positions_per_second": 9436.00629759305,
+      "required_pages_per_second_upper_estimate": 0.0037641639929763245
     },
     {
       "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": 18957.00010270384,
-      "required_pages_per_second_upper_estimate": 0.004385368341953667
+      "required_token_positions_per_second": 18872.0125951861,
+      "required_pages_per_second_upper_estimate": 0.004365707977818485
     },
     {
       "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": 28435.50015405576,
-      "required_pages_per_second_upper_estimate": 0.0017315727698426625
+      "required_token_positions_per_second": 28308.01889277915,
+      "required_pages_per_second_upper_estimate": 0.0017238098298734041
     },
     {
       "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": 28435.50015405576,
-      "required_pages_per_second_upper_estimate": 0.002189043516361591
+      "required_token_positions_per_second": 28308.01889277915,
+      "required_pages_per_second_upper_estimate": 0.0021792296559778003
     },
     {
       "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": 28435.50015405576,
-      "required_pages_per_second_upper_estimate": 0.35307374441630257
+      "required_token_positions_per_second": 28308.01889277915,
+      "required_pages_per_second_upper_estimate": 0.3514908538035828
     }
   ],
   "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": 0.06809770096865009,
-    "producer_p95_seconds": 0.23291494300065096,
-    "gpu_update_seconds": 0.972305739312152,
-    "producer_batches_per_second": 14.684783565018835,
-    "gpu_batches_per_second": 1.0284830784887067,
-    "producer_headroom": 14.278099340824578,
+    "producer_batch_seconds": 0.11615368088274636,
+    "producer_p95_seconds": 0.27200758599792607,
+    "gpu_update_seconds": 0.9766843841923707,
+    "producer_batches_per_second": 8.6092837730168,
+    "gpu_batches_per_second": 1.023872211110357,
+    "producer_headroom": 8.408553020186284,
     "sustainable": true,
-    "buffer_seconds": 60.28295583735343,
+    "buffer_seconds": 60.55443181992698,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 310758.9024081364,
-    "upload_bytes_per_second": 3278642.7063048533,
+    "generation_bytes_per_second": 287792.0535520739,
+    "upload_bytes_per_second": 3094975.390778224,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.68162635977811,
+    "estimated_full_shards_per_hour": 61.75347523614562,
     "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": 1.0451152109999384,
-      "bytes": 1099454,
-      "pages": 31,
-      "shards": 1
-    },
     {
       "seconds": 1.1156518739990133,
       "bytes": 7370564,
@@ -562,6 +556,12 @@
       "bytes": 1169016,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 12.285631268001453,
+      "bytes": 12643645,
+      "pages": 42,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -151,3 +151,4 @@
 {"step": 295300, "loss": 0.6048957770690322, "lr": 0.0003, "grad_norm": 0.16730942682486583, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43223904710, "tokens_per_second": 149659.93292838745, "data_wait_s": 1.556899951538071e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 295400, "loss": 0.586178058758378, "lr": 0.0003, "grad_norm": 0.1444518754760649, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43238613010, "tokens_per_second": 150779.14276032688, "data_wait_s": 8.979000995168462e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 295500, "loss": 0.5683505292981863, "lr": 0.0003, "grad_norm": 0.15291408250282582, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43253262817, "tokens_per_second": 149585.33872645616, "data_wait_s": 1.1811996955657378e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 295600, "loss": 0.5909181497991085, "lr": 0.0003, "grad_norm": 0.15410086744153798, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43268035802, "tokens_per_second": 151579.548667534, "data_wait_s": 1.327800055150874e-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:c0f82d5863c172047e4caa9d98979f562006056667d0373e8990f232971fa540
-size 326568
+oid sha256:0727f9c762c6b912a074e7a551cd618521836e2d95480d3998d8df151fd09664
+size 328686