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": 76000.9208451847,
-      "required_pages_per_second_upper_estimate": 0.005257920050502035
+      "required_token_positions_per_second": 76061.41339155668,
+      "required_pages_per_second_upper_estimate": 0.005262105065222095
     },
     {
       "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": 9500.115105648087,
-      "required_pages_per_second_upper_estimate": 0.003977499809772342
+      "required_token_positions_per_second": 9507.676673944585,
+      "required_pages_per_second_upper_estimate": 0.0039806656805145835
     },
     {
       "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": 19000.230211296173,
-      "required_pages_per_second_upper_estimate": 0.003957246883246839
+      "required_token_positions_per_second": 19015.35334788917,
+      "required_pages_per_second_upper_estimate": 0.003960396633775228
     },
     {
       "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": 28500.345316944262,
-      "required_pages_per_second_upper_estimate": 0.0017085521842432549
+      "required_token_positions_per_second": 28523.030021833754,
+      "required_pages_per_second_upper_estimate": 0.0017099120976638413
     },
     {
       "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": 28500.345316944262,
-      "required_pages_per_second_upper_estimate": 0.0022157764886217963
+      "required_token_positions_per_second": 28523.030021833754,
+      "required_pages_per_second_upper_estimate": 0.0022175401246474825
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28500.345316944262,
-      "required_pages_per_second_upper_estimate": 0.365571828439146
+      "required_token_positions_per_second": 28523.030021833754,
+      "required_pages_per_second_upper_estimate": 0.36586280347653
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09162468057789397,
-    "producer_p95_seconds": 0.2665032049990259,
-    "gpu_update_seconds": 0.9700935091324133,
-    "producer_batches_per_second": 10.914089890331004,
-    "gpu_batches_per_second": 1.0308284619843844,
-    "producer_headroom": 10.587687760697799,
+    "producer_batch_seconds": 0.4888679152577424,
+    "producer_p95_seconds": 0.27726509899730445,
+    "gpu_update_seconds": 0.9693219822310625,
+    "producer_batches_per_second": 2.0455423004653044,
+    "gpu_batches_per_second": 1.0316489446554453,
+    "producer_headroom": 1.9827891174245167,
     "sustainable": true,
-    "buffer_seconds": 60.14579756620962,
+    "buffer_seconds": 60.097962898325875,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 476016.42458789697,
+    "generation_bytes_per_second": 418138.05875493854,
     "upload_bytes_per_second": 5222746.663379439,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 102.14204362132723,
+    "estimated_full_shards_per_hour": 89.72269365297429,
     "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-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -229,3 +229,4 @@
 {"step": 742900, "loss": 0.5111637115478516, "lr": 5e-05, "grad_norm": 0.22492677924003393, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108740399764, "tokens_per_second": 149454.1128898666, "data_wait_s": 1.6869002138264477e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 743000, "loss": 0.5352374315261841, "lr": 5e-05, "grad_norm": 0.22055568912012133, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108754935247, "tokens_per_second": 148846.01957060315, "data_wait_s": 1.2898999557364732e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 743100, "loss": 0.5451949238777161, "lr": 5e-05, "grad_norm": 0.2491799781659081, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108769514755, "tokens_per_second": 150268.03400213382, "data_wait_s": 1.3527998817153275e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 743200, "loss": 0.5074306130409241, "lr": 5e-05, "grad_norm": 0.22568622470124966, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108784215551, "tokens_per_second": 151115.52954239104, "data_wait_s": 1.3770000805379823e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:405329e8bddfec3dc54ecbb7081ef3f2ba7489de53bae19616cefe2b936ff97f
-size 492555
+oid sha256:ba662c8bfec4a61556319e7fc113d9e58ae87cce23de0e6f630fbc50b5b822dd
+size 494673