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": 73811.33271822303,
-      "required_pages_per_second_upper_estimate": 0.006397779246903103
+      "required_token_positions_per_second": 74828.68057773677,
+      "required_pages_per_second_upper_estimate": 0.006485960380921178
     },
     {
       "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": 9226.416589777878,
-      "required_pages_per_second_upper_estimate": 0.0036805555248834686
+      "required_token_positions_per_second": 9353.585072217096,
+      "required_pages_per_second_upper_estimate": 0.003731284933866721
     },
     {
       "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": 18452.833179555757,
-      "required_pages_per_second_upper_estimate": 0.004268738197318153
+      "required_token_positions_per_second": 18707.170144434192,
+      "required_pages_per_second_upper_estimate": 0.004327574577965078
     },
     {
       "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": 27679.249769333634,
-      "required_pages_per_second_upper_estimate": 0.0016855210891522107
+      "required_token_positions_per_second": 28060.755216651287,
+      "required_pages_per_second_upper_estimate": 0.001708752769289466
     },
     {
       "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": 27679.249769333634,
-      "required_pages_per_second_upper_estimate": 0.0021308252683106313
+      "required_token_positions_per_second": 28060.755216651287,
+      "required_pages_per_second_upper_estimate": 0.002160194613719814
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27679.249769333634,
-      "required_pages_per_second_upper_estimate": 0.3436836456452765
+      "required_token_positions_per_second": 28060.755216651287,
+      "required_pages_per_second_upper_estimate": 0.34842066648436476
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.1058975355543339,
-    "producer_p95_seconds": 0.35098780800035456,
-    "gpu_update_seconds": 0.9988710037448971,
-    "producer_batches_per_second": 9.443090386998856,
-    "gpu_batches_per_second": 1.0011302723283289,
-    "producer_headroom": 9.432429173315336,
+    "producer_batch_seconds": 0.08422958879674525,
+    "producer_p95_seconds": 0.2999559260024398,
+    "gpu_update_seconds": 0.9852906590195278,
+    "producer_batches_per_second": 11.872312500695022,
+    "gpu_batches_per_second": 1.0149289357874454,
+    "producer_headroom": 11.697678607895577,
     "sustainable": true,
-    "buffer_seconds": 60.93113122843872,
+    "buffer_seconds": 60.102730200191196,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 499505.0627058349,
-    "upload_bytes_per_second": 5830005.945304935,
+    "generation_bytes_per_second": 500002.9608828235,
+    "upload_bytes_per_second": 5772717.936289432,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 107.18215857392583,
+    "estimated_full_shards_per_hour": 107.28899593223123,
     "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": 1.0802435129990045,
-      "bytes": 11583568,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 1.2433872929977952,
       "bytes": 7696542,
@@ -562,6 +556,12 @@
       "bytes": 2450943,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 0.8343555079991347,
+      "bytes": 5999582,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -255,3 +255,4 @@
 {"step": 25500, "loss": 0.6764032393693924, "lr": 0.0003, "grad_norm": 0.1288349939310712, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3731302372, "tokens_per_second": 147038.4015671184, "data_wait_s": 1.3329001376405358e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 25600, "loss": 0.6448036432266235, "lr": 0.0003, "grad_norm": 0.14769247642991706, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3745867477, "tokens_per_second": 148652.01603643238, "data_wait_s": 1.3130000297678635e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 25700, "loss": 0.6349320039153099, "lr": 0.0003, "grad_norm": 0.14398465931087587, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3760551023, "tokens_per_second": 148601.50113188717, "data_wait_s": 1.2879001587862149e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 25800, "loss": 0.6204517334699631, "lr": 0.0003, "grad_norm": 0.13723740660540482, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3775161785, "tokens_per_second": 148692.41507626904, "data_wait_s": 7.5120005931239575e-06, "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 sha256:7e5c2aa3973ca97ae5c5a84ebceb4d16b2aaee77b57f8ea09504f56890d6da41
-size 542632
+oid sha256:0e0940fcd63d8e5994610059e35efd5e1f0b3852ffc3748139267fd43fdaae78
+size 544750