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": 74700.3282418678,
-      "required_pages_per_second_upper_estimate": 0.006474835125754076
+      "required_token_positions_per_second": 74788.26603654472,
+      "required_pages_per_second_upper_estimate": 0.00648245735092037
     },
     {
       "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": 9337.541030233475,
-      "required_pages_per_second_upper_estimate": 0.0037248847256396503
+      "required_token_positions_per_second": 9348.53325456809,
+      "required_pages_per_second_upper_estimate": 0.0037292696882751278
     },
     {
       "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": 18675.08206046695,
-      "required_pages_per_second_upper_estimate": 0.004320151564470246
+      "required_token_positions_per_second": 18697.06650913618,
+      "required_pages_per_second_upper_estimate": 0.0043252372797567915
     },
     {
       "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": 28012.623090700425,
-      "required_pages_per_second_upper_estimate": 0.0017058217753488066
+      "required_token_positions_per_second": 28045.59976370427,
+      "required_pages_per_second_upper_estimate": 0.001707829881719509
     },
     {
       "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": 28012.623090700425,
-      "required_pages_per_second_upper_estimate": 0.0021564892694258646
+      "required_token_positions_per_second": 28045.59976370427,
+      "required_pages_per_second_upper_estimate": 0.0021590279049989738
     },
     {
       "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": 28012.623090700425,
-      "required_pages_per_second_upper_estimate": 0.34782302656791814
+      "required_token_positions_per_second": 28045.59976370427,
+      "required_pages_per_second_upper_estimate": 0.348232486480801
     }
   ],
   "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": 0.17529349749253242,
-    "producer_p95_seconds": 0.35974394799995935,
-    "gpu_update_seconds": 0.9869836148682031,
-    "producer_batches_per_second": 5.704718168696477,
-    "gpu_batches_per_second": 1.0131880458152642,
-    "producer_headroom": 5.630463359944364,
+    "producer_batch_seconds": 0.11814339372676841,
+    "producer_p95_seconds": 0.33968102899962105,
+    "gpu_update_seconds": 0.9858230964196091,
+    "producer_batches_per_second": 8.464290456330648,
+    "gpu_batches_per_second": 1.014380778490461,
+    "producer_headroom": 8.344293026654826,
     "sustainable": true,
-    "buffer_seconds": 60.206000506960386,
+    "buffer_seconds": 60.13520888159616,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 334417.6995763937,
-    "upload_bytes_per_second": 6046170.251260843,
+    "generation_bytes_per_second": 334726.6386802902,
+    "upload_bytes_per_second": 6002120.100189317,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.75825348347529,
+    "estimated_full_shards_per_hour": 71.8245446234372,
     "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": 1.2663461060001282,
-      "bytes": 11989707,
-      "pages": 36,
-      "shards": 4
-    },
     {
       "seconds": 0.9998438940019696,
       "bytes": 6675780,
@@ -562,6 +556,12 @@
       "bytes": 5854630,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.0005343860029825,
+      "bytes": 7274137,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -284,3 +284,4 @@
 {"step": 113400, "loss": 0.6115688942372799, "lr": 0.0003, "grad_norm": 0.13229779074199938, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16593919775, "tokens_per_second": 146876.21633626384, "data_wait_s": 1.2367003364488482e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 113500, "loss": 0.659650132060051, "lr": 0.0003, "grad_norm": 0.14444335248204354, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16608529611, "tokens_per_second": 148318.61076288426, "data_wait_s": 1.6183999832719564e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 113600, "loss": 0.5900808796286583, "lr": 0.0003, "grad_norm": 0.13183782730153024, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16623051619, "tokens_per_second": 147391.6169124183, "data_wait_s": 1.0732997907325625e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 113700, "loss": 0.5845879912376404, "lr": 0.0003, "grad_norm": 0.13359613735129647, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16637808603, "tokens_per_second": 150483.86248958623, "data_wait_s": 1.4690005627926439e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3acabdf6817b31f6531c268c661b9e03868c46024aa984030b9d0d072e28a4cd
-size 609828
+oid sha256:7cc9498989b7cc661278ae0d30631cb2d0f6a3d66a3a18cb5f3174c6bf109d33
+size 611946