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": 75782.23104236458,
-      "required_pages_per_second_upper_estimate": 0.005242790582512687
+      "required_token_positions_per_second": 76111.78554615434,
+      "required_pages_per_second_upper_estimate": 0.005265589927756662
     },
     {
       "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": 9472.778880295573,
-      "required_pages_per_second_upper_estimate": 0.003966054703062542
+      "required_token_positions_per_second": 9513.973193269292,
+      "required_pages_per_second_upper_estimate": 0.0039833019016695635
     },
     {
       "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": 18945.557760591146,
-      "required_pages_per_second_upper_estimate": 0.003945860053574464
+      "required_token_positions_per_second": 19027.946386538584,
+      "required_pages_per_second_upper_estimate": 0.003963019431625139
     },
     {
       "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": 28418.33664088672,
-      "required_pages_per_second_upper_estimate": 0.001703635889333603
+      "required_token_positions_per_second": 28541.919579807876,
+      "required_pages_per_second_upper_estimate": 0.001711044497821705
     },
     {
       "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": 28418.33664088672,
-      "required_pages_per_second_upper_estimate": 0.002209400674776366
+      "required_token_positions_per_second": 28541.919579807876,
+      "required_pages_per_second_upper_estimate": 0.0022190087046935858
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28418.33664088672,
-      "required_pages_per_second_upper_estimate": 0.36451990919673577
+      "required_token_positions_per_second": 28541.919579807876,
+      "required_pages_per_second_upper_estimate": 0.36610509844419487
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07537969735938077,
-    "producer_p95_seconds": 0.13564731899987237,
-    "gpu_update_seconds": 0.9728929722164527,
-    "producer_batches_per_second": 13.266171595680373,
-    "gpu_batches_per_second": 1.0278622916987383,
-    "producer_headroom": 12.906565113654958,
+    "producer_batch_seconds": 0.06261783455465064,
+    "producer_p95_seconds": 0.13589905800017732,
+    "gpu_update_seconds": 0.9686804674328813,
+    "producer_batches_per_second": 15.969891119873129,
+    "gpu_batches_per_second": 1.0323321607280047,
+    "producer_headroom": 15.469721594850922,
     "sustainable": true,
-    "buffer_seconds": 60.319364277420064,
+    "buffer_seconds": 60.058188980838636,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 440689.4478448507,
+    "generation_bytes_per_second": 459035.4098344091,
     "upload_bytes_per_second": 5527774.183076838,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.56169678219929,
+    "estimated_full_shards_per_hour": 98.4983131530209,
     "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-v1/1790911926322536473-f9bff804/train.jsonl

@@ -18,3 +18,4 @@
 {"step": 485000, "loss": 0.5554909706115723, "lr": 0.0003, "grad_norm": 0.1641477165417348, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70987550848, "tokens_per_second": 152536.30470855438, "data_wait_s": 9.10300013856613e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 485100, "loss": 0.5988941788673401, "lr": 0.0003, "grad_norm": 0.17389446853095353, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71002266879, "tokens_per_second": 139459.88657819288, "data_wait_s": 9.113000032812124e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 485200, "loss": 0.5575228929519653, "lr": 0.0003, "grad_norm": 0.15014661432860948, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71016926460, "tokens_per_second": 152317.7148232644, "data_wait_s": 1.313599977947888e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 485300, "loss": 0.5298693180084229, "lr": 0.0003, "grad_norm": 0.16530683468196739, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71031548750, "tokens_per_second": 152159.4869173255, "data_wait_s": 8.695999895280693e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790911926.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.444.0.1790911926322536473-f9bff804

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e9a30ca85827db114912a64a496ff3b459d4829218246c842926f7011a830d28
-size 43308
+oid sha256:f1faa6e02ccb826ecf4e22b6619970b7010ad7abefd8aa22cdb248d449b9e66a
+size 45426