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": 75519.48462696973,
-      "required_pages_per_second_upper_estimate": 0.0065458375250818344
+      "required_token_positions_per_second": 75894.75671130214,
+      "required_pages_per_second_upper_estimate": 0.006578365158233366
     },
     {
       "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": 9439.935578371216,
-      "required_pages_per_second_upper_estimate": 0.003765731441826718
+      "required_token_positions_per_second": 9486.844588912767,
+      "required_pages_per_second_upper_estimate": 0.0037844441474839505
     },
     {
       "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": 18879.871156742433,
-      "required_pages_per_second_upper_estimate": 0.004367525917728582
+      "required_token_positions_per_second": 18973.689177825534,
+      "required_pages_per_second_upper_estimate": 0.004389229065765381
     },
     {
       "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": 28319.806735113652,
-      "required_pages_per_second_upper_estimate": 0.0017245276476255463
+      "required_token_positions_per_second": 28460.533766738303,
+      "required_pages_per_second_upper_estimate": 0.0017330971855138142
     },
     {
       "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": 28319.806735113652,
-      "required_pages_per_second_upper_estimate": 0.0021801371167115466
+      "required_token_positions_per_second": 28460.533766738303,
+      "required_pages_per_second_upper_estimate": 0.0021909706731633683
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28319.806735113652,
-      "required_pages_per_second_upper_estimate": 0.3516372193540069
+      "required_token_positions_per_second": 28460.533766738303,
+      "required_pages_per_second_upper_estimate": 0.35338457810370766
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "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,
+    "producer_batch_seconds": 0.28763443817969403,
+    "producer_p95_seconds": 0.17811750500004564,
+    "gpu_update_seconds": 0.9714505085042394,
+    "producer_batches_per_second": 3.476635156515123,
+    "gpu_batches_per_second": 1.029388518762236,
+    "producer_headroom": 3.377378990680332,
     "sustainable": true,
-    "buffer_seconds": 60.52922663044158,
+    "buffer_seconds": 60.22993152726284,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 354023.2726203027,
+    "generation_bytes_per_second": 367213.6076867729,
     "upload_bytes_per_second": 3995897.515568216,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.9651530643159,
+    "estimated_full_shards_per_hour": 78.7954919142951,
     "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-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -41,3 +41,4 @@
 {"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}
+{"step": 4400, "loss": 0.8444898128509521, "lr": 0.0003, "grad_norm": 0.19591136392072275, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 643440986, "tokens_per_second": 149340.0880277449, "data_wait_s": 1.2088999937986955e-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 sha256:8070b5364fc4abae503843ccfc365eaae40496d2436269ccf1dfbee24b7e9bb5
-size 89695
+oid sha256:e8e8a940eddc094966ef8b2fe5db51a5f148230eb3793024258074bb08b78769
+size 91778