Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77404.14761090375,
-      "required_pages_per_second_upper_estimate": 0.005354998534089155
+      "required_token_positions_per_second": 77415.95746076122,
+      "required_pages_per_second_upper_estimate": 0.005355815566904922
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9675.518451362968,
-      "required_pages_per_second_upper_estimate": 0.004050937527784789
+      "required_token_positions_per_second": 9676.994682595152,
+      "required_pages_per_second_upper_estimate": 0.004051555594974491
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19351.036902725937,
-      "required_pages_per_second_upper_estimate": 0.0040303106656770765
+      "required_token_positions_per_second": 19353.989365190304,
+      "required_pages_per_second_upper_estimate": 0.004030925585746733
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29026.555354088905,
-      "required_pages_per_second_upper_estimate": 0.0017400976724938723
+      "required_token_positions_per_second": 29030.984047785456,
+      "required_pages_per_second_upper_estimate": 0.0017403631659187568
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29026.555354088905,
-      "required_pages_per_second_upper_estimate": 0.002256687004456448
+      "required_token_positions_per_second": 29030.984047785456,
+      "required_pages_per_second_upper_estimate": 0.0022570313159115903
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29026.555354088905,
-      "required_pages_per_second_upper_estimate": 0.37232148579531954
+      "required_token_positions_per_second": 29030.984047785456,
+      "required_pages_per_second_upper_estimate": 0.37237829232289804
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1985873933905964,
-    "producer_p95_seconds": 0.17393877599988627,
-    "gpu_update_seconds": 0.9525070978188009,
-    "producer_batches_per_second": 5.035566371693726,
-    "gpu_batches_per_second": 1.0498609430732386,
-    "producer_headroom": 4.79641271057594,
+    "producer_batch_seconds": 0.087694478828098,
+    "producer_p95_seconds": 0.1691764229999535,
+    "gpu_update_seconds": 0.9523617923006316,
+    "producer_batches_per_second": 11.403226444394948,
+    "gpu_batches_per_second": 1.0500211244135365,
+    "producer_headroom": 10.85999717459393,
     "sustainable": true,
-    "buffer_seconds": 60.00794716258446,
+    "buffer_seconds": 60.95115470724042,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 498639.9596485408,
-    "upload_bytes_per_second": 2129636.277989188,
+    "generation_bytes_per_second": 518628.9758046245,
+    "upload_bytes_per_second": 2241587.39208421,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 106.996527596399,
+    "estimated_full_shards_per_hour": 111.28570514301349,
     "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": [
@@ -262,6 +262,12 @@
       "bytes": 6044241,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.1359251849999055,
+      "bytes": 7093211,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -58,3 +58,4 @@
 {"step": 690800, "loss": 0.5475606918334961, "lr": 8e-05, "grad_norm": 0.20118566788462452, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101114401085, "tokens_per_second": 153311.38541503355, "data_wait_s": 1.3629000022774562e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 690900, "loss": 0.5649669170379639, "lr": 8e-05, "grad_norm": 0.20676005263549074, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101129081222, "tokens_per_second": 153118.6356530891, "data_wait_s": 1.3066999599686824e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 691000, "loss": 0.5859728455543518, "lr": 8e-05, "grad_norm": 0.19990079891753568, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101143788337, "tokens_per_second": 153881.91135526047, "data_wait_s": 1.6629999663564377e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 691100, "loss": 0.5225908756256104, "lr": 8e-05, "grad_norm": 0.19654221319883927, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101158552772, "tokens_per_second": 153460.82393507086, "data_wait_s": 1.6150999726960436e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dfb741a869cf46d204c4992cdc7293e891d3b36a853d4c86177239c382030e00
-size 128028
+oid sha256:f5f04e1915ae98478c4a2d355230d5c697e147c5c273e6742691519804b9760d
+size 130146