Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77226.53618998604,
-      "required_pages_per_second_upper_estimate": 0.005342710963874794
+      "required_token_positions_per_second": 77233.14402465195,
+      "required_pages_per_second_upper_estimate": 0.00534316810920927
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9653.317023748255,
-      "required_pages_per_second_upper_estimate": 0.004041642253661037
+      "required_token_positions_per_second": 9654.143003081494,
+      "required_pages_per_second_upper_estimate": 0.004041988073959454
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19306.63404749651,
-      "required_pages_per_second_upper_estimate": 0.0040210627219148245
+      "required_token_positions_per_second": 19308.286006162987,
+      "required_pages_per_second_upper_estimate": 0.004021406781339972
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28959.951071244766,
-      "required_pages_per_second_upper_estimate": 0.0017361048474362176
+      "required_token_positions_per_second": 28962.429009244483,
+      "required_pages_per_second_upper_estimate": 0.0017362533960357082
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28959.951071244766,
-      "required_pages_per_second_upper_estimate": 0.0022515088144265958
+      "required_token_positions_per_second": 28962.429009244483,
+      "required_pages_per_second_upper_estimate": 0.002251701463206776
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28959.951071244766,
-      "required_pages_per_second_upper_estimate": 0.3714671575691021
+      "required_token_positions_per_second": 28962.429009244483,
+      "required_pages_per_second_upper_estimate": 0.3714989418971599
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07792152542184994,
-    "producer_p95_seconds": 0.20778168099968752,
-    "gpu_update_seconds": 0.9546977455860606,
-    "producer_batches_per_second": 12.833424327696624,
-    "gpu_batches_per_second": 1.0474519340004618,
-    "producer_headroom": 12.252041273801273,
+    "producer_batch_seconds": 0.0973851105859751,
+    "producer_p95_seconds": 0.18070698799965612,
+    "gpu_update_seconds": 0.9546160645288096,
+    "producer_batches_per_second": 10.268510185827266,
+    "gpu_batches_per_second": 1.0475415584940857,
+    "producer_headroom": 9.80248478216842,
     "sustainable": true,
-    "buffer_seconds": 60.14595797192182,
+    "buffer_seconds": 60.14081206531501,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 483751.7066383607,
-    "upload_bytes_per_second": 3296481.0432165996,
+    "generation_bytes_per_second": 506388.59442510357,
+    "upload_bytes_per_second": 3362898.916958673,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.80185508120647,
+    "estimated_full_shards_per_hour": 108.65920424046355,
     "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": [
@@ -370,6 +370,12 @@
       "bytes": 6174564,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.00537013399844,
+      "bytes": 7625189,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -80,3 +80,4 @@
 {"step": 693000, "loss": 0.5250784158706665, "lr": 8e-05, "grad_norm": 0.20792641863577016, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101436770099, "tokens_per_second": 152125.52605987774, "data_wait_s": 1.6616999346297234e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 693100, "loss": 0.5646812915802002, "lr": 8e-05, "grad_norm": 0.21472038339092928, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101451418741, "tokens_per_second": 152606.0290865611, "data_wait_s": 1.799100027710665e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 693200, "loss": 0.5479682087898254, "lr": 8e-05, "grad_norm": 0.20277307574670675, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101465960696, "tokens_per_second": 151566.30594759306, "data_wait_s": 1.572499968460761e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 693300, "loss": 0.5431538820266724, "lr": 8e-05, "grad_norm": 0.21509999129922094, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101480654045, "tokens_per_second": 153770.85615508325, "data_wait_s": 1.5404999430757016e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2894c2ab4f53201374d632a11ba8304904fd2340f7370523abe5cd42084bd258
-size 174624
+oid sha256:ec3270225a408dc14137747abfe1227acb7fb4a326c84cf9de149c2cf0775fbc
+size 176742