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": 77042.37411100448,
-      "required_pages_per_second_upper_estimate": 0.005329970203935954
+      "required_token_positions_per_second": 77483.12759747294,
+      "required_pages_per_second_upper_estimate": 0.00536046255281883
     },
     {
       "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": 9630.29676387556,
-      "required_pages_per_second_upper_estimate": 0.00403200415157003
+      "required_token_positions_per_second": 9685.390949684117,
+      "required_pages_per_second_upper_estimate": 0.004055070936670646
     },
     {
       "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": 19260.59352775112,
-      "required_pages_per_second_upper_estimate": 0.004011473695822014
+      "required_token_positions_per_second": 19370.781899368234,
+      "required_pages_per_second_upper_estimate": 0.004034423027767089
     },
     {
       "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": 28890.89029162668,
-      "required_pages_per_second_upper_estimate": 0.001731964759147818
+      "required_token_positions_per_second": 29056.17284905235,
+      "required_pages_per_second_upper_estimate": 0.0017418731960936345
     },
     {
       "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": 28890.89029162668,
-      "required_pages_per_second_upper_estimate": 0.002246139642577556
+      "required_token_positions_per_second": 29056.17284905235,
+      "required_pages_per_second_upper_estimate": 0.0022589896344162633
     },
     {
       "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": 28890.89029162668,
-      "required_pages_per_second_upper_estimate": 0.37058132003984917
+      "required_token_positions_per_second": 29056.17284905235,
+      "required_pages_per_second_upper_estimate": 0.37270138721992213
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08503996153123694,
-    "producer_p95_seconds": 0.23329554100018868,
-    "gpu_update_seconds": 0.9569798549272501,
-    "producer_batches_per_second": 11.759177473671354,
-    "gpu_batches_per_second": 1.0449540759413585,
-    "producer_headroom": 11.253295952817801,
+    "producer_batch_seconds": 0.07890590552344179,
+    "producer_p95_seconds": 0.20752425200043945,
+    "gpu_update_seconds": 0.951536189698215,
+    "producer_batches_per_second": 12.673322653941467,
+    "gpu_batches_per_second": 1.0509321777000995,
+    "producer_headroom": 12.059125148947533,
     "sustainable": true,
-    "buffer_seconds": 60.28973086041676,
+    "buffer_seconds": 60.89831614068576,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 535051.025969339,
-    "upload_bytes_per_second": 5382216.631376611,
+    "generation_bytes_per_second": 534952.3714766715,
+    "upload_bytes_per_second": 5372392.324397347,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 114.80949482259871,
+    "estimated_full_shards_per_hour": 114.78832586503134,
     "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": 1.483484325000063,
-      "bytes": 6080107,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.1274267459998555,
       "bytes": 2324297,
@@ -562,6 +556,12 @@
       "bytes": 11876531,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 0.9798814419991686,
+      "bytes": 2581068,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -112,3 +112,4 @@
 {"step": 896200, "loss": 0.4921562373638153, "lr": 3e-05, "grad_norm": 0.3187196371497609, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131178647311, "tokens_per_second": 153083.34460836468, "data_wait_s": 1.4336999811348505e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 896300, "loss": 0.5767093300819397, "lr": 3e-05, "grad_norm": 0.5584751470497801, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131193360263, "tokens_per_second": 153426.67742808667, "data_wait_s": 1.7170999853988178e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 896400, "loss": 0.52915358543396, "lr": 3e-05, "grad_norm": 0.31634236585155595, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131207877377, "tokens_per_second": 152045.3253269531, "data_wait_s": 1.336299828835763e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 896500, "loss": 0.5308711528778076, "lr": 3e-05, "grad_norm": 0.31997333905983366, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131222365870, "tokens_per_second": 151356.98767846482, "data_wait_s": 1.1264999557170086e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3060fc2708911ec077de812d498427b0500fb99bff9072718a268b5966c73869
-size 243183
+oid sha256:bd2d209fec1113639fa1b847f8b2d49ebced6ffacf0ac2269ec45a1f15046d6c
+size 245301