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": 76074.90388292362,
-      "required_pages_per_second_upper_estimate": 0.006593979858490684
+      "required_token_positions_per_second": 76048.14775092195,
+      "required_pages_per_second_upper_estimate": 0.006591660704781576
     },
     {
       "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": 9509.362985365453,
-      "required_pages_per_second_upper_estimate": 0.003793427072508957
+      "required_token_positions_per_second": 9506.018468865244,
+      "required_pages_per_second_upper_estimate": 0.0037920928948720458
     },
     {
       "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": 19018.725970730906,
-      "required_pages_per_second_upper_estimate": 0.004399647535183554
+      "required_token_positions_per_second": 19012.03693773049,
+      "required_pages_per_second_upper_estimate": 0.004398100145121868
     },
     {
       "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": 28528.08895609636,
-      "required_pages_per_second_upper_estimate": 0.0017372109421110334
+      "required_token_positions_per_second": 28518.055406595733,
+      "required_pages_per_second_upper_estimate": 0.0017365999515884113
     },
     {
       "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": 28528.08895609636,
-      "required_pages_per_second_upper_estimate": 0.0021961712586448844
+      "required_token_positions_per_second": 28518.055406595733,
+      "required_pages_per_second_upper_estimate": 0.002195398848229684
     },
     {
       "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": 28528.08895609636,
-      "required_pages_per_second_upper_estimate": 0.35422338746286003
+      "required_token_positions_per_second": 28518.055406595733,
+      "required_pages_per_second_upper_estimate": 0.3540988043581923
     }
   ],
   "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": 0.05854420603128574,
-    "producer_p95_seconds": 0.20118910500059428,
-    "gpu_update_seconds": 0.9691500907245914,
-    "producer_batches_per_second": 17.081109605715806,
-    "gpu_batches_per_second": 1.0318319211551057,
-    "producer_headroom": 16.55415892405616,
+    "producer_batch_seconds": 0.08230423897659023,
+    "producer_p95_seconds": 0.20416940600080125,
+    "gpu_update_seconds": 0.9694910682306023,
+    "producer_batches_per_second": 12.150042481826844,
+    "gpu_batches_per_second": 1.031469017889024,
+    "producer_headroom": 11.779357664753505,
     "sustainable": true,
-    "buffer_seconds": 60.08730562492467,
+    "buffer_seconds": 60.10844623029734,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 457113.29167530424,
-    "upload_bytes_per_second": 6343065.379659409,
+    "generation_bytes_per_second": 486236.14728622825,
+    "upload_bytes_per_second": 6216230.369070896,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.08587134069772,
+    "estimated_full_shards_per_hour": 104.33495820942055,
     "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": 1.532430633000331,
-      "bytes": 13601194,
-      "pages": 37,
-      "shards": 4
-    },
     {
       "seconds": 1.1585651320001489,
       "bytes": 6030177,
@@ -562,6 +556,12 @@
       "bytes": 7415356,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 0.9046226490008848,
+      "bytes": 1040439,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -142,3 +142,4 @@
 {"step": 54200, "loss": 0.6311823390424252, "lr": 0.0003, "grad_norm": 0.13768664016934543, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7932586156, "tokens_per_second": 150525.70052595058, "data_wait_s": 1.546299972687848e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 54300, "loss": 0.5882640983909369, "lr": 0.0003, "grad_norm": 0.12783607859267374, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7947334095, "tokens_per_second": 151531.96730554063, "data_wait_s": 1.4601000657421537e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 54400, "loss": 0.6611931398510933, "lr": 0.0003, "grad_norm": 0.1324707736323593, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7962099279, "tokens_per_second": 151628.78406672596, "data_wait_s": 7.717999324086122e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 54500, "loss": 0.6466125007718801, "lr": 0.0003, "grad_norm": 0.13557349814196135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7976756607, "tokens_per_second": 150395.21269945215, "data_wait_s": 8.86999987415038e-06, "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:ddc0832ebd9a25f66abd02b40614c2fd3b50d0040156ca0bf162e8454c9b723c
-size 306723
+oid sha256:c9559161bb728dedf70ae3a2c1d3ba9590ccf2b93e359e85d73667133f61dad6
+size 308841