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": 77213.67016974345,
-      "required_pages_per_second_upper_estimate": 0.005341820862741132
+      "required_token_positions_per_second": 76983.27565663037,
+      "required_pages_per_second_upper_estimate": 0.005325881635735046
     },
     {
       "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": 9651.708771217931,
-      "required_pages_per_second_upper_estimate": 0.004040968911910722
+      "required_token_positions_per_second": 9622.909457078797,
+      "required_pages_per_second_upper_estimate": 0.004028911240478733
     },
     {
       "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": 19303.417542435862,
-      "required_pages_per_second_upper_estimate": 0.004020392808735633
+      "required_token_positions_per_second": 19245.818914157593,
+      "required_pages_per_second_upper_estimate": 0.0040083965334432444
     },
     {
       "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": 28955.126313653793,
-      "required_pages_per_second_upper_estimate": 0.001735815610585617
+      "required_token_positions_per_second": 28868.72837123639,
+      "required_pages_per_second_upper_estimate": 0.0017306361858597133
     },
     {
       "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": 28955.126313653793,
-      "required_pages_per_second_upper_estimate": 0.0022511337107457632
+      "required_token_positions_per_second": 28868.72837123639,
+      "required_pages_per_second_upper_estimate": 0.0022444166507471965
     },
     {
       "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": 28955.126313653793,
-      "required_pages_per_second_upper_estimate": 0.37140527075914614
+      "required_token_positions_per_second": 28868.72837123639,
+      "required_pages_per_second_upper_estimate": 0.3702970507206987
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.06715492870314321,
-    "producer_p95_seconds": 0.15040887099985412,
-    "gpu_update_seconds": 0.9548568257138834,
-    "producer_batches_per_second": 14.890939791187577,
-    "gpu_batches_per_second": 1.0472774274325012,
-    "producer_headroom": 14.218715500909928,
+    "producer_batch_seconds": 0.0937514389062244,
+    "producer_p95_seconds": 0.14647329400031595,
+    "gpu_update_seconds": 0.9577145083933565,
+    "producer_batches_per_second": 10.666502953626747,
+    "gpu_batches_per_second": 1.0441525018531679,
+    "producer_headroom": 10.215464632508924,
     "sustainable": true,
-    "buffer_seconds": 60.155980019974656,
+    "buffer_seconds": 60.336014028781456,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 406325.31236660416,
-    "upload_bytes_per_second": 5339466.466482653,
+    "generation_bytes_per_second": 408372.1552414754,
+    "upload_bytes_per_second": 5023638.975944563,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.18795326469987,
+    "estimated_full_shards_per_hour": 87.62715809758373,
     "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": [
@@ -268,6 +268,12 @@
       "bytes": 6013324,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 2.719455530999767,
+      "bytes": 7662793,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -19,3 +19,4 @@
 {"step": 442100, "loss": 0.5817098021507263, "lr": 0.0003, "grad_norm": 0.16716236427496123, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64710338440, "tokens_per_second": 153906.83277553265, "data_wait_s": 1.580200023454381e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 442200, "loss": 0.5698598623275757, "lr": 0.0003, "grad_norm": 0.16258972882849176, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64725016724, "tokens_per_second": 152293.02992094893, "data_wait_s": 1.3707000107388012e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 442300, "loss": 0.5957845449447632, "lr": 0.0003, "grad_norm": 0.15658732256581287, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64739600490, "tokens_per_second": 151319.5431445865, "data_wait_s": 1.513600000180304e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 442400, "loss": 0.5786598920822144, "lr": 0.0003, "grad_norm": 0.15098705925138126, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64754049223, "tokens_per_second": 149183.2007289611, "data_wait_s": 1.3804999980493449e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2f6bbd463cf74cbd6a2386a48cd6730f4086d12149a400dccfb36d535c9411ca
-size 44643
+oid sha256:fe16ca1d97cd6c03f52550d45c347ea57b28891857e1342594fc6c2ebef039c9
+size 46761