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": 76106.21886174838,
-      "required_pages_per_second_upper_estimate": 0.006596694161488125
+      "required_token_positions_per_second": 75610.00152368422,
+      "required_pages_per_second_upper_estimate": 0.006553683300276069
     },
     {
       "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": 9513.277357718547,
-      "required_pages_per_second_upper_estimate": 0.003794988574165688
+      "required_token_positions_per_second": 9451.250190460527,
+      "required_pages_per_second_upper_estimate": 0.00377024500975767
     },
     {
       "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": 19026.554715437094,
-      "required_pages_per_second_upper_estimate": 0.00440145857749013
+      "required_token_positions_per_second": 18902.500380921054,
+      "required_pages_per_second_upper_estimate": 0.004372760790481563
     },
     {
       "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": 28539.83207315564,
-      "required_pages_per_second_upper_estimate": 0.0017379260363285702
+      "required_token_positions_per_second": 28353.750571381577,
+      "required_pages_per_second_upper_estimate": 0.0017265946491647176
     },
     {
       "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": 28539.83207315564,
-      "required_pages_per_second_upper_estimate": 0.0021970752763031368
+      "required_token_positions_per_second": 28353.750571381577,
+      "required_pages_per_second_upper_estimate": 0.002182750207190004
     },
     {
       "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": 28539.83207315564,
-      "required_pages_per_second_upper_estimate": 0.3543691976750517
+      "required_token_positions_per_second": 28353.750571381577,
+      "required_pages_per_second_upper_estimate": 0.3520586881977424
     }
   ],
   "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": 0.3049812858828176,
-    "producer_p95_seconds": 0.19523727799969492,
-    "gpu_update_seconds": 0.9687513202295787,
-    "producer_batches_per_second": 3.2788897099221628,
-    "gpu_batches_per_second": 1.0322566577385577,
-    "producer_headroom": 3.1764287353742757,
+    "producer_batch_seconds": 0.46107704274221106,
+    "producer_p95_seconds": 0.18859210000027815,
+    "gpu_update_seconds": 0.9751090929009611,
+    "producer_batches_per_second": 2.1688349392817234,
+    "gpu_batches_per_second": 1.0255262793468454,
+    "producer_headroom": 2.1148506702949126,
     "sustainable": true,
-    "buffer_seconds": 60.062581854233876,
+    "buffer_seconds": 60.45676375985959,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 342694.02452333865,
-    "upload_bytes_per_second": 5695977.380152808,
+    "generation_bytes_per_second": 308881.54077139765,
+    "upload_bytes_per_second": 5699859.105253986,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.53416015410538,
+    "estimated_full_shards_per_hour": 66.27878825527617,
     "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": 7217811,
       "pages": 25,
       "shards": 3
+    },
+    {
+      "seconds": 0.9584536009997464,
+      "bytes": 5627627,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -71,3 +71,4 @@
 {"step": 47100, "loss": 0.7052925955504179, "lr": 0.0003, "grad_norm": 0.13154949091454196, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6892270079, "tokens_per_second": 149512.43425594695, "data_wait_s": 8.834000254864804e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 47200, "loss": 0.6300609912723303, "lr": 0.0003, "grad_norm": 0.1442390169924938, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6906944627, "tokens_per_second": 149643.27268759083, "data_wait_s": 8.23199934529839e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 47300, "loss": 0.6709001772105694, "lr": 0.0003, "grad_norm": 0.1646743449629838, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6921637712, "tokens_per_second": 137226.13754580484, "data_wait_s": 8.320000233652536e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 47400, "loss": 0.6294107064604759, "lr": 0.0003, "grad_norm": 0.12717438974419965, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6936350677, "tokens_per_second": 150994.34081723812, "data_wait_s": 9.824000699154567e-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:eea84c0a50adac3657bb50b93ebf8b96bb0318f251acf5a18b62aab5f5609aa6
-size 155562
+oid sha256:43f7f9515c6ed7de93e082849883f3f39861a6858d2cc7a9d36c7fa16ef623bf
+size 157680