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": 75991.66298552987,
-      "required_pages_per_second_upper_estimate": 0.006586764748476718
+      "required_token_positions_per_second": 75706.16289318418,
+      "required_pages_per_second_upper_estimate": 0.006562018324065564
     },
     {
       "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": 9498.957873191233,
-      "required_pages_per_second_upper_estimate": 0.00378927631769237
+      "required_token_positions_per_second": 9463.270361648023,
+      "required_pages_per_second_upper_estimate": 0.0037750400357619367
     },
     {
       "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": 18997.915746382467,
-      "required_pages_per_second_upper_estimate": 0.004394833456027983
+      "required_token_positions_per_second": 18926.540723296046,
+      "required_pages_per_second_upper_estimate": 0.004378322100594442
     },
     {
       "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": 28496.8736195737,
-      "required_pages_per_second_upper_estimate": 0.0017353100918910173
+      "required_token_positions_per_second": 28389.811084944067,
+      "required_pages_per_second_upper_estimate": 0.0017287905452457802
     },
     {
       "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": 28496.8736195737,
-      "required_pages_per_second_upper_estimate": 0.0021937682156297884
+      "required_token_positions_per_second": 28389.811084944067,
+      "required_pages_per_second_upper_estimate": 0.002185526245345935
     },
     {
       "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": 28496.8736195737,
-      "required_pages_per_second_upper_estimate": 0.3538357974542595
+      "required_token_positions_per_second": 28389.811084944067,
+      "required_pages_per_second_upper_estimate": 0.35250643908941315
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.2599786846017196,
-    "producer_p95_seconds": 0.3278312980000919,
-    "gpu_update_seconds": 0.9702116930121544,
-    "producer_batches_per_second": 3.846469188548951,
-    "gpu_batches_per_second": 1.0307028942264793,
-    "producer_headroom": 3.7318893835411653,
+    "producer_batch_seconds": 0.09770617866411158,
+    "producer_p95_seconds": 0.2913737779999792,
+    "gpu_update_seconds": 0.9738705170413243,
+    "producer_batches_per_second": 10.23476727544263,
+    "gpu_batches_per_second": 1.0268305513940996,
+    "producer_headroom": 9.96733809833294,
     "sustainable": true,
-    "buffer_seconds": 60.15312496675357,
+    "buffer_seconds": 60.37997205656211,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 158,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 551522.8514715539,
-    "upload_bytes_per_second": 5086695.112074043,
+    "generation_bytes_per_second": 574688.1191770971,
+    "upload_bytes_per_second": 5223660.530144832,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 118.34396513090098,
+    "estimated_full_shards_per_hour": 61.657346160338804,
     "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": 1.0936733489998005,
-      "bytes": 876493,
-      "pages": 25,
-      "shards": 1
-    },
     {
       "seconds": 1.364482594999572,
       "bytes": 12283477,
@@ -562,6 +556,12 @@
       "bytes": 6230072,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.0601522930010105,
+      "bytes": 12072218,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -237,3 +237,4 @@
 {"step": 63700, "loss": 0.69142622128129, "lr": 0.0003, "grad_norm": 0.13678521866561938, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9322635639, "tokens_per_second": 149529.36993722356, "data_wait_s": 1.541299934615381e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 63800, "loss": 0.6414812821894884, "lr": 0.0003, "grad_norm": 0.13486914322527982, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9337318391, "tokens_per_second": 150186.49814084015, "data_wait_s": 1.0626001312630251e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 63900, "loss": 0.6261835917830467, "lr": 0.0003, "grad_norm": 0.13677463701096396, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9351811507, "tokens_per_second": 147353.44496967868, "data_wait_s": 1.4470002497546375e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 64000, "loss": 0.6302631683647633, "lr": 0.0003, "grad_norm": 0.13780771526704533, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9366397123, "tokens_per_second": 148638.90808606657, "data_wait_s": 1.1954001820413396e-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 sha256:0c50b763c080295a4e8b9e1ee36c4c439540e7f02bdd03983e29c50ceed4ea0e
-size 509499
+oid sha256:4d795b0dea2517ab300cc4921a6b6389b8e2641665d102e51ffae640567a9bd1
+size 511617