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": 77243.52897191232,
-      "required_pages_per_second_upper_estimate": 0.0053438865639571405
+      "required_token_positions_per_second": 77016.5503774188,
+      "required_pages_per_second_upper_estimate": 0.005328183658127179
     },
     {
       "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": 9655.44112148904,
-      "required_pages_per_second_upper_estimate": 0.004042531569028899
+      "required_token_positions_per_second": 9627.06879717735,
+      "required_pages_per_second_upper_estimate": 0.004030652669321099
     },
     {
       "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": 19310.88224297808,
-      "required_pages_per_second_upper_estimate": 0.004021947509001186
+      "required_token_positions_per_second": 19254.1375943547,
+      "required_pages_per_second_upper_estimate": 0.0040101290951499525
     },
     {
       "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": 28966.32336446712,
-      "required_pages_per_second_upper_estimate": 0.0017364868566849693
+      "required_token_positions_per_second": 28881.206391532047,
+      "required_pages_per_second_upper_estimate": 0.001731384224123604
     },
     {
       "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": 28966.32336446712,
-      "required_pages_per_second_upper_estimate": 0.0022520042321958777
+      "required_token_positions_per_second": 28881.206391532047,
+      "required_pages_per_second_upper_estimate": 0.002245386761940173
     },
     {
       "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": 28966.32336446712,
-      "required_pages_per_second_upper_estimate": 0.37154889450452305
+      "required_token_positions_per_second": 28881.206391532047,
+      "required_pages_per_second_upper_estimate": 0.370457105367197
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08146737680468163,
-    "producer_p95_seconds": 0.20548267399863107,
-    "gpu_update_seconds": 0.9544877219010715,
-    "producer_batches_per_second": 12.274852084626513,
-    "gpu_batches_per_second": 1.0476824133560156,
-    "producer_headroom": 11.716195602927778,
+    "producer_batch_seconds": 0.1009426049687363,
+    "producer_p95_seconds": 0.19742387599944777,
+    "gpu_update_seconds": 0.9573007313193945,
+    "producer_batches_per_second": 9.906619710376184,
+    "gpu_batches_per_second": 1.0446038191381672,
+    "producer_headroom": 9.48361429364625,
     "sustainable": true,
-    "buffer_seconds": 60.1327264797675,
+    "buffer_seconds": 60.30994607312186,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 538573.3712625764,
-    "upload_bytes_per_second": 5811879.521093396,
+    "generation_bytes_per_second": 556773.396089745,
+    "upload_bytes_per_second": 5709887.759243697,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 115.5653081265256,
+    "estimated_full_shards_per_hour": 119.4706097795416,
     "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": 1.1878640209997684,
-      "bytes": 11776096,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.224845865000134,
       "bytes": 2065038,
@@ -562,6 +556,12 @@
       "bytes": 5802765,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.9111343770000531,
+      "bytes": 2649643,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -126,3 +126,4 @@
 {"step": 537600, "loss": 0.5897620916366577, "lr": 0.0003, "grad_norm": 0.18373614094457494, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78687009323, "tokens_per_second": 152701.0595191887, "data_wait_s": 1.4770001143915579e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 537700, "loss": 0.5979024767875671, "lr": 0.0003, "grad_norm": 0.18848251933706925, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78701551851, "tokens_per_second": 152424.60191982138, "data_wait_s": 1.3967999620945193e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 537800, "loss": 0.622713565826416, "lr": 0.0003, "grad_norm": 0.19116680577474718, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78716370523, "tokens_per_second": 155401.68683552614, "data_wait_s": 1.0193998605245724e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 537900, "loss": 0.5968229174613953, "lr": 0.0003, "grad_norm": 0.18029057161348447, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78731126659, "tokens_per_second": 153842.39206109423, "data_wait_s": 1.4650000593974255e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3f1ce6851283c090aed459722cbe83d991de4adef7069dc42bd729ae75a8d5b6
-size 272835
+oid sha256:294814e1eab178b16ded357c1cd5dc79506fed2555a284f89a00e74b543e4d83
+size 274953