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": 76585.05866512343,
-      "required_pages_per_second_upper_estimate": 0.005298332060271875
+      "required_token_positions_per_second": 76861.91240049498,
+      "required_pages_per_second_upper_estimate": 0.005317485444074049
     },
     {
       "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": 9573.13233314043,
-      "required_pages_per_second_upper_estimate": 0.004008070598150288
+      "required_token_positions_per_second": 9607.739050061873,
+      "required_pages_per_second_upper_estimate": 0.0040225597078548695
     },
     {
       "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": 19146.26466628086,
-      "required_pages_per_second_upper_estimate": 0.003987662009032589
+      "required_token_positions_per_second": 19215.478100123746,
+      "required_pages_per_second_upper_estimate": 0.0040020773420210674
     },
     {
       "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": 28719.396999421286,
-      "required_pages_per_second_upper_estimate": 0.0017216839981352
+      "required_token_positions_per_second": 28823.217150185617,
+      "required_pages_per_second_upper_estimate": 0.0017279078576493303
     },
     {
       "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": 28719.396999421286,
-      "required_pages_per_second_upper_estimate": 0.00223280679342786
+      "required_token_positions_per_second": 28823.217150185617,
+      "required_pages_per_second_upper_estimate": 0.0022408783534932047
     },
     {
       "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": 28719.396999421286,
-      "required_pages_per_second_upper_estimate": 0.3683815882225893
+      "required_token_positions_per_second": 28823.217150185617,
+      "required_pages_per_second_upper_estimate": 0.36971328164320133
     }
   ],
   "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": 0.44947936374225606,
-    "producer_p95_seconds": 0.2724706179978966,
-    "gpu_update_seconds": 0.9626943072849727,
-    "producer_batches_per_second": 2.224796243534392,
-    "gpu_batches_per_second": 1.0387513382313833,
-    "producer_headroom": 2.1417986785195513,
+    "producer_batch_seconds": 0.09695241501580654,
+    "producer_p95_seconds": 0.276220543000818,
+    "gpu_update_seconds": 0.9592267183756047,
+    "producer_batches_per_second": 10.314338222900028,
+    "gpu_batches_per_second": 1.0425064073417831,
+    "producer_headroom": 9.89378880576846,
     "sustainable": true,
-    "buffer_seconds": 60.649741358953285,
+    "buffer_seconds": 60.4312832576631,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 295570.0330834692,
+    "generation_bytes_per_second": 283655.9295470901,
     "upload_bytes_per_second": 4971640.8025979,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.42244858148629,
+    "estimated_full_shards_per_hour": 60.86595930871512,
     "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-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -199,3 +199,4 @@
 {"step": 779900, "loss": 0.521408200263977, "lr": 3e-05, "grad_norm": 0.239432013755575, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114155151564, "tokens_per_second": 151979.67729621593, "data_wait_s": 1.020299896481447e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 780000, "loss": 0.5338215231895447, "lr": 3e-05, "grad_norm": 0.25660716764034813, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114169855740, "tokens_per_second": 151961.01306357514, "data_wait_s": 1.405599687132053e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 780100, "loss": 0.5658770799636841, "lr": 3e-05, "grad_norm": 0.2438217714019965, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114184617151, "tokens_per_second": 145341.33308708438, "data_wait_s": 9.096002031583339e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 780200, "loss": 0.5623610615730286, "lr": 3e-05, "grad_norm": 0.24890262904165042, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114199317450, "tokens_per_second": 152029.68073092971, "data_wait_s": 9.111998224398121e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8b764f5bb2920b146c8e6697dfa1fadb2279793cb759bef0ce3418c0419de540
-size 429015
+oid sha256:5797b1820860eae2ea7e6d00d3c419cffd823bfe7b368a9de707fa97190fd92b
+size 431133