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": 75798.6672275844,
-      "required_pages_per_second_upper_estimate": 0.006570036365321282
+      "required_token_positions_per_second": 75575.13142082456,
+      "required_pages_per_second_upper_estimate": 0.0065506608481376615
     },
     {
       "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": 9474.83340344805,
-      "required_pages_per_second_upper_estimate": 0.003779652706018849
+      "required_token_positions_per_second": 9446.89142760307,
+      "required_pages_per_second_upper_estimate": 0.0037685062340845183
     },
     {
       "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": 18949.6668068961,
-      "required_pages_per_second_upper_estimate": 0.004383671913030145
+      "required_token_positions_per_second": 18893.78285520614,
+      "required_pages_per_second_upper_estimate": 0.00437074414433063
     },
     {
       "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": 28424.500210344155,
-      "required_pages_per_second_upper_estimate": 0.0017309029309828703
+      "required_token_positions_per_second": 28340.674282809206,
+      "required_pages_per_second_upper_estimate": 0.001725798371796644
     },
     {
       "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": 28424.500210344155,
-      "required_pages_per_second_upper_estimate": 0.002188196710244879
+      "required_token_positions_per_second": 28340.674282809206,
+      "required_pages_per_second_upper_estimate": 0.0021817435582982196
     },
     {
       "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": 28424.500210344155,
-      "required_pages_per_second_upper_estimate": 0.3529371619298478
+      "required_token_positions_per_second": 28340.674282809206,
+      "required_pages_per_second_upper_estimate": 0.35189632445719615
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.12504075224234157,
-    "producer_p95_seconds": 0.42566588499903446,
-    "gpu_update_seconds": 0.9726820100758862,
-    "producer_batches_per_second": 7.997392706514588,
-    "gpu_batches_per_second": 1.028085221728304,
-    "producer_headroom": 7.7789200131388405,
+    "producer_batch_seconds": 0.5019751500158804,
+    "producer_p95_seconds": 0.4566120489980676,
+    "gpu_update_seconds": 0.9755590048459303,
+    "producer_batches_per_second": 1.992130486874428,
+    "gpu_batches_per_second": 1.0250533233076247,
+    "producer_headroom": 1.9434408352984556,
     "sustainable": true,
-    "buffer_seconds": 60.306284624704944,
+    "buffer_seconds": 60.48465830044768,
     "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": 326172.4089895161,
+    "generation_bytes_per_second": 297744.9908125559,
     "upload_bytes_per_second": 5278868.57076614,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.98900606407273,
+    "estimated_full_shards_per_hour": 63.88914387972362,
     "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-v1/1790738827394728502-c744ae20/train.jsonl

@@ -399,3 +399,4 @@
 {"step": 354900, "loss": 0.584769681096077, "lr": 0.0003, "grad_norm": 0.1662222847641233, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51947299203, "tokens_per_second": 140361.33777305816, "data_wait_s": 9.903000318445265e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 355000, "loss": 0.5933996718376875, "lr": 0.0003, "grad_norm": 0.14682717746036786, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51961890047, "tokens_per_second": 147682.00578222997, "data_wait_s": 1.4045996067579836e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 355100, "loss": 0.6090170033276081, "lr": 0.0003, "grad_norm": 0.15771364523541026, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51976574318, "tokens_per_second": 116100.11540562313, "data_wait_s": 1.5910001820884645e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 355200, "loss": 0.552221804857254, "lr": 0.0003, "grad_norm": 0.14781253353344473, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51991218815, "tokens_per_second": 148360.2892172157, "data_wait_s": 1.1461997928563505e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:131b5fb40908cfdd8b136c878a6dd5c0b77d7208056f2240e0397cbb81a8447b
-size 855747
+oid sha256:94f2b68db5e7d6d9aebe2b56cfceef8c945342c19dfae183c563cd2d1edee046
+size 857865