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": 77198.30405065721,
-      "required_pages_per_second_upper_estimate": 0.005340757798968437
+      "required_token_positions_per_second": 76585.05866512343,
+      "required_pages_per_second_upper_estimate": 0.005298332060271875
     },
     {
       "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": 9649.788006332152,
-      "required_pages_per_second_upper_estimate": 0.0040401647277631784
+      "required_token_positions_per_second": 9573.13233314043,
+      "required_pages_per_second_upper_estimate": 0.004008070598150288
     },
     {
       "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": 19299.576012664304,
-      "required_pages_per_second_upper_estimate": 0.004019592719392171
+      "required_token_positions_per_second": 19146.26466628086,
+      "required_pages_per_second_upper_estimate": 0.003987662009032589
     },
     {
       "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": 28949.364018996454,
-      "required_pages_per_second_upper_estimate": 0.001735470169819424
+      "required_token_positions_per_second": 28719.396999421286,
+      "required_pages_per_second_upper_estimate": 0.0017216839981352
     },
     {
       "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": 28949.364018996454,
-      "required_pages_per_second_upper_estimate": 0.0022506857176818108
+      "required_token_positions_per_second": 28719.396999421286,
+      "required_pages_per_second_upper_estimate": 0.00223280679342786
     },
     {
       "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": 28949.364018996454,
-      "required_pages_per_second_upper_estimate": 0.37133135823035174
+      "required_token_positions_per_second": 28719.396999421286,
+      "required_pages_per_second_upper_estimate": 0.3683815882225893
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.39113858623451847,
-    "producer_p95_seconds": 0.258143648999976,
-    "gpu_update_seconds": 0.9550468874500143,
-    "producer_batches_per_second": 2.5566385807827743,
-    "gpu_batches_per_second": 1.047069011103749,
-    "producer_headroom": 2.4417097189112105,
+    "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,
     "sustainable": true,
-    "buffer_seconds": 60.1679539093509,
+    "buffer_seconds": 60.649741358953285,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 310654.2999911742,
-    "upload_bytes_per_second": 4935605.042243135,
+    "generation_bytes_per_second": 295570.0330834692,
+    "upload_bytes_per_second": 4971640.8025979,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.65918111611767,
+    "estimated_full_shards_per_hour": 63.42244858148629,
     "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": [
-    {
-      "seconds": 2.3685029429998394,
-      "bytes": 11410659,
-      "pages": 34,
-      "shards": 4
-    },
     {
       "seconds": 0.8315555410008528,
       "bytes": 1017212,
@@ -562,6 +556,12 @@
       "bytes": 13032899,
       "pages": 32,
       "shards": 4
+    },
+    {
+      "seconds": 1.520787964000192,
+      "bytes": 10348218,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -198,3 +198,4 @@
 {"step": 779800, "loss": 0.5133309364318848, "lr": 3e-05, "grad_norm": 0.24772370564504653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114140554045, "tokens_per_second": 148428.1273339138, "data_wait_s": 1.1033000191673636e-05, "gpu_peak_gb": 89.37899398803711}
 {"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}

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 sha256:4d359b8bb528ae59d7014dca6df40981383c4eef82ddbb3e9ab744987b535f6e
-size 426897
+oid sha256:8b764f5bb2920b146c8e6697dfa1fadb2279793cb759bef0ce3418c0419de540
+size 429015