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": 75682.06925783536,
-      "required_pages_per_second_upper_estimate": 0.006559929948818289
+      "required_token_positions_per_second": 75682.7077236738,
+      "required_pages_per_second_upper_estimate": 0.006559985289418987
     },
     {
       "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": 9460.25865722942,
-      "required_pages_per_second_upper_estimate": 0.0037738386218403625
+      "required_token_positions_per_second": 9460.338465459225,
+      "required_pages_per_second_upper_estimate": 0.0037738704585364704
     },
     {
       "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": 18920.51731445884,
-      "required_pages_per_second_upper_estimate": 0.00437692869096833
+      "required_token_positions_per_second": 18920.67693091845,
+      "required_pages_per_second_upper_estimate": 0.004376965615427107
     },
     {
       "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": 28380.77597168826,
-      "required_pages_per_second_upper_estimate": 0.0017282403542520774
+      "required_token_positions_per_second": 28381.015396377676,
+      "required_pages_per_second_upper_estimate": 0.0017282549339594983
     },
     {
       "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": 28380.77597168826,
-      "required_pages_per_second_upper_estimate": 0.0021848306973167087
+      "required_token_positions_per_second": 28381.015396377676,
+      "required_pages_per_second_upper_estimate": 0.0021848491288920698
     },
     {
       "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": 28380.77597168826,
-      "required_pages_per_second_upper_estimate": 0.3523942532213549
+      "required_token_positions_per_second": 28381.015396377676,
+      "required_pages_per_second_upper_estimate": 0.35239722607469454
     }
   ],
   "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": 0.08572530486719643,
-    "producer_p95_seconds": 0.24220952299947385,
-    "gpu_update_seconds": 0.9741805519193959,
-    "producer_batches_per_second": 11.665167030308913,
-    "gpu_batches_per_second": 1.0265037605500673,
-    "producer_headroom": 11.363978855818278,
+    "producer_batch_seconds": 0.12235561875763779,
+    "producer_p95_seconds": 0.2672280540027714,
+    "gpu_update_seconds": 0.9741723336483855,
+    "producer_batches_per_second": 8.172898066747566,
+    "gpu_batches_per_second": 1.026512420297225,
+    "producer_headroom": 7.961811182353854,
     "sustainable": true,
-    "buffer_seconds": 60.399194219002545,
+    "buffer_seconds": 60.3986846861999,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 276504.60952663777,
-    "upload_bytes_per_second": 5953268.568310719,
+    "generation_bytes_per_second": 289102.4386608227,
+    "upload_bytes_per_second": 5874373.048434086,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.331452506536,
+    "estimated_full_shards_per_hour": 62.034653376278975,
     "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": 1.030597670998759,
-      "bytes": 7980071,
-      "pages": 26,
-      "shards": 3
-    },
     {
       "seconds": 0.9780282619994978,
       "bytes": 7468851,
@@ -562,6 +556,12 @@
       "bytes": 16530913,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 0.8704905230006261,
+      "bytes": 1186552,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -195,3 +195,4 @@
 {"step": 19500, "loss": 0.6772917918860912, "lr": 0.0003, "grad_norm": 0.136990950203081, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2853181602, "tokens_per_second": 147552.31416240722, "data_wait_s": 1.0497999028302729e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 19600, "loss": 0.6757691092789173, "lr": 0.0003, "grad_norm": 0.1404766323672605, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2867804437, "tokens_per_second": 132248.53525720828, "data_wait_s": 1.001499913400039e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 19700, "loss": 0.6688902676105499, "lr": 0.0003, "grad_norm": 0.14548545177305264, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2882535757, "tokens_per_second": 149313.88675079896, "data_wait_s": 1.1452000762801617e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 19800, "loss": 0.6570432372391224, "lr": 0.0003, "grad_norm": 0.1425050550810938, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2897152642, "tokens_per_second": 148921.66507115713, "data_wait_s": 1.770500239217654e-05, "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 sha256:2ad307f6305df28f4a8f997da6f8c584999ee7cd91c408432f9e69c1f8979285
-size 413986
+oid sha256:d82fcc78294a70188389eb3db21faacb06f485d42c5da6e0f1693b62b3be706d
+size 416104