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": 76937.02385904267,
-      "required_pages_per_second_upper_estimate": 0.0053226818290589735
+      "required_token_positions_per_second": 77122.3085353717,
+      "required_pages_per_second_upper_estimate": 0.005335500252887096
     },
     {
       "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": 9617.127982380334,
-      "required_pages_per_second_upper_estimate": 0.004026490657753407
+      "required_token_positions_per_second": 9640.288566921463,
+      "required_pages_per_second_upper_estimate": 0.004036187510852775
     },
     {
       "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": 19234.25596476067,
-      "required_pages_per_second_upper_estimate": 0.004005988276019347
+      "required_token_positions_per_second": 19280.577133842926,
+      "required_pages_per_second_upper_estimate": 0.004015635753967806
     },
     {
       "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": 28851.383947141003,
-      "required_pages_per_second_upper_estimate": 0.001729596414118597
+      "required_token_positions_per_second": 28920.865700764392,
+      "required_pages_per_second_upper_estimate": 0.0017337617391558243
     },
     {
       "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": 28851.383947141003,
-      "required_pages_per_second_upper_estimate": 0.002243068198063839
+      "required_token_positions_per_second": 28920.865700764392,
+      "required_pages_per_second_upper_estimate": 0.002248470098790122
     },
     {
       "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": 28851.383947141003,
-      "required_pages_per_second_upper_estimate": 0.3700745750713947
+      "required_token_positions_per_second": 28920.865700764392,
+      "required_pages_per_second_upper_estimate": 0.370965812403181
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.5247473010781363,
-    "producer_p95_seconds": 0.20191293500101892,
-    "gpu_update_seconds": 0.9582902522338014,
-    "producer_batches_per_second": 1.9056791677544946,
-    "gpu_batches_per_second": 1.043525171699255,
-    "producer_headroom": 1.8261937703441555,
+    "producer_batch_seconds": 0.11381871815619604,
+    "producer_p95_seconds": 0.21427506300096866,
+    "gpu_update_seconds": 0.9559879806526418,
+    "producer_batches_per_second": 8.785901090782598,
+    "gpu_batches_per_second": 1.0460382559593602,
+    "producer_headroom": 8.399215841991099,
     "sustainable": true,
-    "buffer_seconds": 60.372285890729486,
+    "buffer_seconds": 60.22724278111643,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 74,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 206942.07474815546,
-    "upload_bytes_per_second": 4497303.51048844,
+    "generation_bytes_per_second": 217027.9765285454,
+    "upload_bytes_per_second": 4568907.797879928,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 44.4049518760061,
+    "estimated_full_shards_per_hour": 46.56915161030075,
     "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": 1.0848658659997454,
-      "bytes": 5569114,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 0.9946771380000428,
       "bytes": 7343286,
@@ -562,6 +556,12 @@
       "bytes": 11778517,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.142588306000107,
+      "bytes": 12693549,
+      "pages": 49,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -116,3 +116,4 @@
 {"step": 451800, "loss": 0.5840235948562622, "lr": 0.0003, "grad_norm": 0.15363027850533936, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66129463393, "tokens_per_second": 152427.4422108863, "data_wait_s": 8.916998922359198e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 451900, "loss": 0.5504519939422607, "lr": 0.0003, "grad_norm": 0.15542979430587375, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66144122324, "tokens_per_second": 144212.62021890015, "data_wait_s": 1.4871999155730009e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 452000, "loss": 0.578105628490448, "lr": 0.0003, "grad_norm": 0.17458674679668898, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66158787132, "tokens_per_second": 129967.99176219804, "data_wait_s": 1.2552000043797307e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 452100, "loss": 0.5985599756240845, "lr": 0.0003, "grad_norm": 0.15233343228801152, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66173491630, "tokens_per_second": 153415.9382148165, "data_wait_s": 1.2500000593718141e-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 sha256:6a1d601e3712d5012c7e5f243668a6a846217bd90762a71bded1942dd38ae756
-size 251655
+oid sha256:debd1a68170f5ee873de8280df42f476bfd734a9e94692de83e36a5350010778
+size 253773