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": 76749.27800226526,
-      "required_pages_per_second_upper_estimate": 0.005309693135056706
+      "required_token_positions_per_second": 76863.90993564944,
+      "required_pages_per_second_upper_estimate": 0.0053176236381909546
     },
     {
       "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": 9593.659750283157,
-      "required_pages_per_second_upper_estimate": 0.004016664999046734
+      "required_token_positions_per_second": 9607.98874195618,
+      "required_pages_per_second_upper_estimate": 0.004022664248636856
     },
     {
       "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": 19187.319500566315,
-      "required_pages_per_second_upper_estimate": 0.003996212648325463
+      "required_token_positions_per_second": 19215.97748391236,
+      "required_pages_per_second_upper_estimate": 0.0040021813504946
     },
     {
       "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": 28780.97925084947,
-      "required_pages_per_second_upper_estimate": 0.001725375760077666
+      "required_token_positions_per_second": 28823.96622586854,
+      "required_pages_per_second_upper_estimate": 0.001727952763592749
     },
     {
       "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": 28780.97925084947,
-      "required_pages_per_second_upper_estimate": 0.0022375945426047037
+      "required_token_positions_per_second": 28823.96622586854,
+      "required_pages_per_second_upper_estimate": 0.002240936590832716
     },
     {
       "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": 28780.97925084947,
-      "required_pages_per_second_upper_estimate": 0.3691714992220401
+      "required_token_positions_per_second": 28823.96622586854,
+      "required_pages_per_second_upper_estimate": 0.369722889981767
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07110750154692624,
-    "producer_p95_seconds": 0.17565997099973174,
-    "gpu_update_seconds": 0.9606344439855697,
-    "producer_batches_per_second": 14.06321384165166,
-    "gpu_batches_per_second": 1.0409787055428772,
-    "producer_headroom": 13.509607609425212,
+    "producer_batch_seconds": 0.055326522820287494,
+    "producer_p95_seconds": 0.1788659390003886,
+    "gpu_update_seconds": 0.9592017900432749,
+    "producer_batches_per_second": 18.074513796000087,
+    "gpu_batches_per_second": 1.0425335006462868,
+    "producer_headroom": 17.337105987285153,
     "sustainable": true,
-    "buffer_seconds": 60.519969971090894,
+    "buffer_seconds": 60.42971277272632,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 396398.8413352395,
+    "generation_bytes_per_second": 381142.63917671924,
     "upload_bytes_per_second": 5728801.481405435,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.05796365778816,
+    "estimated_full_shards_per_hour": 81.78433782077964,
     "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-v1/1791039222312944374-63c10d34/train.jsonl

@@ -77,3 +77,4 @@
 {"step": 612100, "loss": 0.5246615409851074, "lr": 0.0001, "grad_norm": 0.1585942307912722, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89593479598, "tokens_per_second": 150958.17484313183, "data_wait_s": 1.0028999895439483e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 612200, "loss": 0.5195173025131226, "lr": 0.0001, "grad_norm": 0.15260576830122513, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89608129676, "tokens_per_second": 152321.39499359208, "data_wait_s": 7.714000275882427e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 612300, "loss": 0.5995924472808838, "lr": 0.0001, "grad_norm": 0.150966046024431, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89622757554, "tokens_per_second": 151422.72025322024, "data_wait_s": 1.694999991741497e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 612400, "loss": 0.5913826823234558, "lr": 0.0001, "grad_norm": 0.1776482835088219, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89637580401, "tokens_per_second": 154143.22949013903, "data_wait_s": 1.5454999811481684e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5d9b3a65bfa298524fb7ec0b0eee1df655da9015bf03408b97a7638a86514609
-size 169053
+oid sha256:1840fa9f64a1ba18ec60b69ef72f0f518eaeb8d8daeea0a27ebc8ae55bb3a3b1
+size 171171