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": 76461.1487181764,
-      "required_pages_per_second_upper_estimate": 0.00528975968263139
+      "required_token_positions_per_second": 74803.89907337657,
+      "required_pages_per_second_upper_estimate": 0.005175107306855177
     },
     {
       "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": 9557.64358977205,
-      "required_pages_per_second_upper_estimate": 0.004001585784743689
+      "required_token_positions_per_second": 9350.487384172071,
+      "required_pages_per_second_upper_estimate": 0.003914853807372467
     },
     {
       "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": 19115.2871795441,
-      "required_pages_per_second_upper_estimate": 0.003981210215476584
+      "required_token_positions_per_second": 18700.974768344142,
+      "required_pages_per_second_upper_estimate": 0.003894919866376668
     },
     {
       "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": 28672.93076931615,
-      "required_pages_per_second_upper_estimate": 0.0017188984185902223
+      "required_token_positions_per_second": 28051.462152516215,
+      "required_pages_per_second_upper_estimate": 0.0016816423239406995
     },
     {
       "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": 28672.93076931615,
-      "required_pages_per_second_upper_estimate": 0.002229194248420532
+      "required_token_positions_per_second": 28051.462152516215,
+      "required_pages_per_second_upper_estimate": 0.0021808777970158916
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28672.93076931615,
-      "required_pages_per_second_upper_estimate": 0.3677855693143514
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28051.462152516215,
+      "required_pages_per_second_upper_estimate": 0.3598140371790538
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11037881891400048,
-    "producer_p95_seconds": 0.22509415599961358,
-    "gpu_update_seconds": 0.9642544120249834,
-    "producer_batches_per_second": 9.059709189125593,
-    "gpu_batches_per_second": 1.0370707020151964,
-    "producer_headroom": 8.735864557277639,
+    "producer_batch_seconds": 0.11511429296099607,
+    "producer_p95_seconds": 0.21724887600066722,
+    "gpu_update_seconds": 0.9856170722822724,
+    "producer_batches_per_second": 8.687018564574148,
+    "gpu_batches_per_second": 1.0145928151228376,
+    "producer_headroom": 8.56207380447732,
     "sustainable": true,
-    "buffer_seconds": 60.74802795757395,
+    "buffer_seconds": 60.122641409218616,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 370532.2816858428,
+    "generation_bytes_per_second": 382146.90524753416,
     "upload_bytes_per_second": 5583115.460778908,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.5076020997187,
+    "estimated_full_shards_per_hour": 81.99982994145888,
     "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": [

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -136,3 +136,4 @@
 {"step": 773600, "loss": 0.5211269855499268, "lr": 3e-05, "grad_norm": 0.26197575174297194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113232832701, "tokens_per_second": 153378.65506054234, "data_wait_s": 1.3980999938212335e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 773700, "loss": 0.5062357783317566, "lr": 3e-05, "grad_norm": 0.2501352750979994, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113247577265, "tokens_per_second": 153189.7574557602, "data_wait_s": 1.0011000995291397e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 773800, "loss": 0.538447380065918, "lr": 3e-05, "grad_norm": 0.25514714318845866, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113262328204, "tokens_per_second": 152890.15194719922, "data_wait_s": 9.745001079863869e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 773900, "loss": 0.5280709266662598, "lr": 3e-05, "grad_norm": 0.2468482830649305, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113276930616, "tokens_per_second": 151849.07020083806, "data_wait_s": 7.816001016180962e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:870822801b149a2381d61cabed52ef4073c3ce6b6b886b40151ce819f5783501
-size 294015
+oid sha256:de1b2e15c00a4b419e0326dcdd7d503d2f5c920b9d51ce4eac88086d3e2217fc
+size 296133