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": 74754.67220544236,
-      "required_pages_per_second_upper_estimate": 0.006479545522783203
+      "required_token_positions_per_second": 75519.48462696973,
+      "required_pages_per_second_upper_estimate": 0.0065458375250818344
     },
     {
       "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": 9344.334025680295,
-      "required_pages_per_second_upper_estimate": 0.003727594553087719
+      "required_token_positions_per_second": 9439.935578371216,
+      "required_pages_per_second_upper_estimate": 0.003765731441826718
     },
     {
       "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": 18688.66805136059,
-      "required_pages_per_second_upper_estimate": 0.004323294444358217
+      "required_token_positions_per_second": 18879.871156742433,
+      "required_pages_per_second_upper_estimate": 0.004367525917728582
     },
     {
       "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": 28033.002077040885,
-      "required_pages_per_second_upper_estimate": 0.0017070627486966625
+      "required_token_positions_per_second": 28319.806735113652,
+      "required_pages_per_second_upper_estimate": 0.0017245276476255463
     },
     {
       "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": 28033.002077040885,
-      "required_pages_per_second_upper_estimate": 0.0021580581002069977
+      "required_token_positions_per_second": 28319.806735113652,
+      "required_pages_per_second_upper_estimate": 0.0021801371167115466
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 28033.002077040885,
-      "required_pages_per_second_upper_estimate": 0.34807606537418684
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28319.806735113652,
+      "required_pages_per_second_upper_estimate": 0.3516372193540069
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10338281624220258,
-    "producer_p95_seconds": 0.1903229330000613,
-    "gpu_update_seconds": 0.986266113205328,
-    "producer_batches_per_second": 9.672787377519548,
-    "gpu_batches_per_second": 1.0139251329948238,
-    "producer_headroom": 9.539942410687761,
+    "producer_batch_seconds": 0.20570541037501755,
+    "producer_p95_seconds": 0.18646120199991856,
+    "gpu_update_seconds": 0.97627784887809,
+    "producer_batches_per_second": 4.861320847988001,
+    "gpu_batches_per_second": 1.0242985653614602,
+    "producer_headroom": 4.745999860179937,
     "sustainable": true,
-    "buffer_seconds": 60.16223290552501,
+    "buffer_seconds": 60.52922663044158,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 385787.7395268363,
-    "upload_bytes_per_second": 3922124.6977980672,
+    "generation_bytes_per_second": 354023.2726203027,
+    "upload_bytes_per_second": 3995897.515568216,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.78106822351282,
+    "estimated_full_shards_per_hour": 75.9651530643159,
     "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": [
@@ -406,6 +406,12 @@
       "bytes": 1092240,
       "pages": 31,
       "shards": 1
+    },
+    {
+      "seconds": 0.9772284549999313,
+      "bytes": 8526326,
+      "pages": 38,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -40,3 +40,4 @@
 {"step": 4000, "loss": 0.781925618648529, "lr": 0.0003, "grad_norm": 0.19247554432262717, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 585075237, "tokens_per_second": 147804.51803401762, "data_wait_s": 9.609000699128956e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 4100, "loss": 0.7929353713989258, "lr": 0.0003, "grad_norm": 0.18971049866724163, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 599597978, "tokens_per_second": 147626.81771538057, "data_wait_s": 1.4173000636219513e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 4200, "loss": 0.8014660682529211, "lr": 0.0003, "grad_norm": 0.18963374644769673, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 614164749, "tokens_per_second": 147722.69298432808, "data_wait_s": 1.2866000361100305e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 4300, "loss": 0.792757298797369, "lr": 0.0003, "grad_norm": 0.19787410055599836, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 628794841, "tokens_per_second": 149029.03684519258, "data_wait_s": 1.2385000445647165e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:94f5263b69f048b6938048d1f8172ef6033633e30f80ff95b073af59a35d8f6c
-size 87612
+oid sha256:8070b5364fc4abae503843ccfc365eaae40496d2436269ccf1dfbee24b7e9bb5
+size 89695