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": 74322.72807271754,
-      "required_pages_per_second_upper_estimate": 0.006442105700111011
+      "required_token_positions_per_second": 75415.08999091137,
+      "required_pages_per_second_upper_estimate": 0.006536788862614082
     },
     {
       "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": 9290.341009089692,
-      "required_pages_per_second_upper_estimate": 0.003706055931502191
+      "required_token_positions_per_second": 9426.88624886392,
+      "required_pages_per_second_upper_estimate": 0.003760525869181395
     },
     {
       "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": 18580.682018179385,
-      "required_pages_per_second_upper_estimate": 0.004298313776071019
+      "required_token_positions_per_second": 18853.77249772784,
+      "required_pages_per_second_upper_estimate": 0.004361488452286268
     },
     {
       "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": 27871.023027269075,
-      "required_pages_per_second_upper_estimate": 0.0016971990744039545
+      "required_token_positions_per_second": 28280.658746591766,
+      "required_pages_per_second_upper_estimate": 0.0017221437405181854
     },
     {
       "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": 27871.023027269075,
-      "required_pages_per_second_upper_estimate": 0.0021455885045688562
+      "required_token_positions_per_second": 28280.658746591766,
+      "required_pages_per_second_upper_estimate": 0.0021771233961865636
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 27871.023027269075,
-      "required_pages_per_second_upper_estimate": 0.34606482768502767
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28280.658746591766,
+      "required_pages_per_second_upper_estimate": 0.35115113235645434
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.07633068497651863,
-    "producer_p95_seconds": 0.18561911399956443,
-    "gpu_update_seconds": 0.9919980322555484,
-    "producer_batches_per_second": 13.100891211805934,
-    "gpu_batches_per_second": 1.0080665157432391,
-    "producer_headroom": 12.996058302905492,
+    "producer_batch_seconds": 0.19299012628904677,
+    "producer_p95_seconds": 0.17815188500026125,
+    "gpu_update_seconds": 0.9776292782901315,
+    "producer_batches_per_second": 5.18161223700259,
+    "gpu_batches_per_second": 1.0228826224895748,
+    "producer_headroom": 5.065695831640156,
     "sustainable": true,
-    "buffer_seconds": 60.51187996758845,
+    "buffer_seconds": 60.61301525398815,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 465184.91522892204,
-    "upload_bytes_per_second": 5445680.559367966,
+    "generation_bytes_per_second": 451844.0056701848,
+    "upload_bytes_per_second": 5469244.7405096935,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.81785385752436,
+    "estimated_full_shards_per_hour": 96.95520522669943,
     "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": [
@@ -340,6 +340,12 @@
       "bytes": 1124293,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.2218557999995028,
+      "bytes": 7398978,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -57,3 +57,4 @@
 {"step": 360700, "loss": 0.5598017778247595, "lr": 0.0003, "grad_norm": 0.14913583663535313, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52796455511, "tokens_per_second": 149662.12951439363, "data_wait_s": 1.2286999663047027e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 360800, "loss": 0.5656123664230108, "lr": 0.0003, "grad_norm": 0.16475022846952875, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52811025956, "tokens_per_second": 148490.95252830585, "data_wait_s": 1.0054000085801817e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 360900, "loss": 0.5878585204482079, "lr": 0.0003, "grad_norm": 0.15594778002520077, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52825694224, "tokens_per_second": 149461.72327852427, "data_wait_s": 8.394000360567588e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 361000, "loss": 0.5639312844723463, "lr": 0.0003, "grad_norm": 0.16376991511278852, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52840419589, "tokens_per_second": 150113.4355381505, "data_wait_s": 1.2329999663052149e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:249f8f57123b38c6ba3dac3ea54f4263179e9a0e4bc07b25671c62c8ea7b1522
-size 125910
+oid sha256:c25cfe2ba7c37d789d21cf2aa528e79322acc2bc32f171e1125949189fa6fb52
+size 128028