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": 76264.98978184494,
-      "required_pages_per_second_upper_estimate": 0.005276188952264541
+      "required_token_positions_per_second": 76668.07115655989,
+      "required_pages_per_second_upper_estimate": 0.005304075057044966
     },
     {
       "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": 9533.123722730617,
-      "required_pages_per_second_upper_estimate": 0.00399131982844649
+      "required_token_positions_per_second": 9583.508894569986,
+      "required_pages_per_second_upper_estimate": 0.0040124150477336
     },
     {
       "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": 19066.247445461235,
-      "required_pages_per_second_upper_estimate": 0.003970996532131879
+      "required_token_positions_per_second": 19167.017789139973,
+      "required_pages_per_second_upper_estimate": 0.0039919843372274895
     },
     {
       "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": 28599.371168191854,
-      "required_pages_per_second_upper_estimate": 0.0017144886328218283
+      "required_token_positions_per_second": 28750.52668370996,
+      "required_pages_per_second_upper_estimate": 0.0017235501751760303
     },
     {
       "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": 28599.371168191854,
-      "required_pages_per_second_upper_estimate": 0.002223475312987609
+      "required_token_positions_per_second": 28750.52668370996,
+      "required_pages_per_second_upper_estimate": 0.0022352269894562937
     },
     {
       "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": 28599.371168191854,
-      "required_pages_per_second_upper_estimate": 0.366842025733275
+      "required_token_positions_per_second": 28750.52668370996,
+      "required_pages_per_second_upper_estimate": 0.36878088638819356
     }
   ],
   "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": 0.08603590903129543,
-    "producer_p95_seconds": 0.2571906310004124,
-    "gpu_update_seconds": 0.9667345424276333,
-    "producer_batches_per_second": 11.623053806943,
-    "gpu_batches_per_second": 1.0344101261643466,
-    "producer_headroom": 11.236407603666803,
+    "producer_batch_seconds": 0.12090996267971832,
+    "producer_p95_seconds": 0.2556093420025718,
+    "gpu_update_seconds": 0.9616519482985803,
+    "producer_batches_per_second": 8.27061705948026,
+    "gpu_batches_per_second": 1.0398772672059446,
+    "producer_headroom": 7.953455008880668,
     "sustainable": true,
-    "buffer_seconds": 60.904276172940904,
+    "buffer_seconds": 60.584072742810555,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 288891.82672342635,
-    "upload_bytes_per_second": 3918586.5120159346,
+    "generation_bytes_per_second": 302289.50252832315,
+    "upload_bytes_per_second": 3931534.4499917794,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.9894609573087,
+    "estimated_full_shards_per_hour": 64.86429030310889,
     "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": 2.011558195001271,
-      "bytes": 5842602,
-      "pages": 41,
-      "shards": 2
-    },
     {
       "seconds": 1.6693870859999151,
       "bytes": 7612861,
@@ -562,6 +556,12 @@
       "bytes": 6116166,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.7439238019978802,
+      "bytes": 6171251,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -215,3 +215,4 @@
 {"step": 821500, "loss": 0.43291744589805603, "lr": 3e-05, "grad_norm": 0.2694788604155506, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120244409128, "tokens_per_second": 150553.77588452472, "data_wait_s": 1.385100040351972e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 821600, "loss": 0.5205039381980896, "lr": 3e-05, "grad_norm": 0.27924293684598367, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120259063634, "tokens_per_second": 151785.38149482908, "data_wait_s": 1.5546000213362277e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 821700, "loss": 0.5221646428108215, "lr": 3e-05, "grad_norm": 0.2711142156545151, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120273731526, "tokens_per_second": 151479.0486378975, "data_wait_s": 9.331000910606235e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 821800, "loss": 0.4260617196559906, "lr": 3e-05, "grad_norm": 0.27389021874056513, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120288473389, "tokens_per_second": 129560.19725595282, "data_wait_s": 1.0712999937823042e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:354f864d769eeead2bc97f1fb56cf27c7d6645c1d6f78b7d8fc4e0a17ae7154c
-size 462903
+oid sha256:1a10892531f76e655e4c926578f441ff65cff030eedd7940e1c7469dff4b0cca
+size 465021