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": 77228.23307714416,
-      "required_pages_per_second_upper_estimate": 0.005342828358465714
+      "required_token_positions_per_second": 77620.05663330329,
+      "required_pages_per_second_upper_estimate": 0.00536993562123671
     },
     {
       "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": 9653.52913464302,
-      "required_pages_per_second_upper_estimate": 0.004041731060063296
+      "required_token_positions_per_second": 9702.507079162911,
+      "required_pages_per_second_upper_estimate": 0.0040622371026579895
     },
     {
       "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": 19307.05826928604,
-      "required_pages_per_second_upper_estimate": 0.0040211510761261025
+      "required_token_positions_per_second": 19405.014158325823,
+      "required_pages_per_second_upper_estimate": 0.004041552704542577
     },
     {
       "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": 28960.587403929065,
-      "required_pages_per_second_upper_estimate": 0.0017361429946090175
+      "required_token_positions_per_second": 29107.521237488734,
+      "required_pages_per_second_upper_estimate": 0.0017449514535759466
     },
     {
       "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": 28960.587403929065,
-      "required_pages_per_second_upper_estimate": 0.002251558286493868
+      "required_token_positions_per_second": 29107.521237488734,
+      "required_pages_per_second_upper_estimate": 0.002262981745759501
     },
     {
       "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": 28960.587403929065,
-      "required_pages_per_second_upper_estimate": 0.3714753197615354
+      "required_token_positions_per_second": 29107.521237488734,
+      "required_pages_per_second_upper_estimate": 0.3733600292131801
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.06668300788287951,
-    "producer_p95_seconds": 0.16936335699938354,
-    "gpu_update_seconds": 0.9546767686158539,
-    "producer_batches_per_second": 14.996324127375551,
-    "gpu_batches_per_second": 1.0474749495055362,
-    "producer_headroom": 14.316642259038856,
+    "producer_batch_seconds": 0.06869286388280926,
+    "producer_p95_seconds": 0.16496308600017073,
+    "gpu_update_seconds": 0.9498575909099068,
+    "producer_batches_per_second": 14.557552902525805,
+    "gpu_batches_per_second": 1.052789396610559,
+    "producer_headroom": 13.827602129536682,
     "sustainable": true,
-    "buffer_seconds": 60.1446364227988,
+    "buffer_seconds": 60.79088581823404,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 328301.61756047775,
+    "generation_bytes_per_second": 340090.58780317067,
     "upload_bytes_per_second": 5873810.8097339235,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.44588465796231,
+    "estimated_full_shards_per_hour": 72.97552323886242,
     "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-v1/1790914434605211575-090606d4/train.jsonl

@@ -32,3 +32,4 @@
 {"step": 488200, "loss": 0.5942641496658325, "lr": 0.0003, "grad_norm": 0.16873086263281123, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71456706571, "tokens_per_second": 153355.76433489085, "data_wait_s": 1.0932000805041753e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 488300, "loss": 0.5797544717788696, "lr": 0.0003, "grad_norm": 0.15258235446268698, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71471202382, "tokens_per_second": 150730.92543615756, "data_wait_s": 1.4066000403545331e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 488400, "loss": 0.5689780116081238, "lr": 0.0003, "grad_norm": 0.17653242025923216, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71485709701, "tokens_per_second": 151669.7361474753, "data_wait_s": 1.0341999768570531e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 488500, "loss": 0.6281889081001282, "lr": 0.0003, "grad_norm": 0.16109204375892366, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71500276570, "tokens_per_second": 151287.73307463256, "data_wait_s": 1.9514000086928718e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:949f88e712dec832cfce23856625ef7d2af75dfbae004569eeb2d6f980215384
-size 72177
+oid sha256:e909c76d845d8edfc54905b323ecddfbf13a945e0e02e6ba848125dc236b6984
+size 74295