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": 76093.86457455029,
-      "required_pages_per_second_upper_estimate": 0.00659562332318545
+      "required_token_positions_per_second": 75965.32004033586,
+      "required_pages_per_second_upper_estimate": 0.006584481408752887
     },
     {
       "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": 9511.733071818786,
-      "required_pages_per_second_upper_estimate": 0.0037943725354311415
+      "required_token_positions_per_second": 9495.665005041983,
+      "required_pages_per_second_upper_estimate": 0.003787962743354868
     },
     {
       "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": 19023.46614363757,
-      "required_pages_per_second_upper_estimate": 0.004400744090761318
+      "required_token_positions_per_second": 18991.330010083966,
+      "required_pages_per_second_upper_estimate": 0.004393309961840325
     },
     {
       "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": 28535.19921545636,
-      "required_pages_per_second_upper_estimate": 0.0017376439196014096
+      "required_token_positions_per_second": 28486.995015125947,
+      "required_pages_per_second_upper_estimate": 0.001734708536709181
     },
     {
       "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": 28535.19921545636,
-      "required_pages_per_second_upper_estimate": 0.002196718626092875
+      "required_token_positions_per_second": 28486.995015125947,
+      "required_pages_per_second_upper_estimate": 0.002193007733313673
     },
     {
       "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": 28535.19921545636,
-      "required_pages_per_second_upper_estimate": 0.354311673087604
+      "required_token_positions_per_second": 28486.995015125947,
+      "required_pages_per_second_upper_estimate": 0.35371313824858075
     }
   ],
   "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": 0.09888203229702697,
-    "producer_p95_seconds": 0.19445026499852247,
-    "gpu_update_seconds": 0.9689086027135286,
-    "producer_batches_per_second": 10.113060753000587,
-    "gpu_batches_per_second": 1.032089091994226,
-    "producer_headroom": 9.798631563346826,
+    "producer_batch_seconds": 0.05444538010165445,
+    "producer_p95_seconds": 0.19827475100100855,
+    "gpu_update_seconds": 0.9705481390830988,
+    "producer_batches_per_second": 18.3670312914137,
+    "gpu_batches_per_second": 1.0303455951651457,
+    "producer_headroom": 17.82608804036261,
     "sustainable": true,
-    "buffer_seconds": 60.07233336823877,
+    "buffer_seconds": 60.17398462315212,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 386978.9080223696,
+    "generation_bytes_per_second": 387119.1468669959,
     "upload_bytes_per_second": 6070616.371807287,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.03666525367085,
+    "estimated_full_shards_per_hour": 83.06675724513443,
     "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-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -96,3 +96,4 @@
 {"step": 49600, "loss": 0.6500741466879845, "lr": 0.0003, "grad_norm": 0.133837719168408, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7258738800, "tokens_per_second": 147919.99334265897, "data_wait_s": 8.693001291248947e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 49700, "loss": 0.6234482005238533, "lr": 0.0003, "grad_norm": 0.14637150633321536, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7273411800, "tokens_per_second": 138581.6965428865, "data_wait_s": 8.326000170200132e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 49800, "loss": 0.5993867069482803, "lr": 0.0003, "grad_norm": 0.13161946458735313, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7288087450, "tokens_per_second": 150009.94783579005, "data_wait_s": 9.265000699087977e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 49900, "loss": 0.6517329514026642, "lr": 0.0003, "grad_norm": 0.13602953573908014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7302896789, "tokens_per_second": 151602.71017741144, "data_wait_s": 1.0196001312579028e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6e82622861d3763a7f9a6268dd01fb3cc2a19850c64ff5c4938c9b483ede1654
-size 208512
+oid sha256:693a1f5a27f108df40fa3842f4ec1616e736bfbe1328bb0ad45cec2bce4abf6c
+size 210630