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": 77025.85316063275,
-      "required_pages_per_second_upper_estimate": 0.005328827246255351
+      "required_token_positions_per_second": 76982.1611879696,
+      "required_pages_per_second_upper_estimate": 0.005325804534207171
     },
     {
       "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": 9628.231645079093,
-      "required_pages_per_second_upper_estimate": 0.004031139529454534
+      "required_token_positions_per_second": 9622.7701484962,
+      "required_pages_per_second_upper_estimate": 0.004028852914884294
     },
     {
       "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": 19256.463290158186,
-      "required_pages_per_second_upper_estimate": 0.0040106134762530965
+      "required_token_positions_per_second": 19245.5402969924,
+      "required_pages_per_second_upper_estimate": 0.004008338504835364
     },
     {
       "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": 28884.694935237283,
-      "required_pages_per_second_upper_estimate": 0.0017315933569920351
+      "required_token_positions_per_second": 28868.3104454886,
+      "required_pages_per_second_upper_estimate": 0.001730611131849269
     },
     {
       "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": 28884.694935237283,
-      "required_pages_per_second_upper_estimate": 0.0022456579808687732
+      "required_token_positions_per_second": 28868.3104454886,
+      "required_pages_per_second_upper_estimate": 0.0022443841588585672
     },
     {
       "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": 28884.694935237283,
-      "required_pages_per_second_upper_estimate": 0.3705018526601414
+      "required_token_positions_per_second": 28868.3104454886,
+      "required_pages_per_second_upper_estimate": 0.37029169001793977
     }
   ],
   "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": 0.07111479046889713,
-    "producer_p95_seconds": 0.2574570759970811,
-    "gpu_update_seconds": 0.9571851135000702,
-    "producer_batches_per_second": 14.061772430270487,
-    "gpu_batches_per_second": 1.0447299962108392,
-    "producer_headroom": 13.459719239680613,
+    "producer_batch_seconds": 0.11740338947637952,
+    "producer_p95_seconds": 0.27001913799904287,
+    "gpu_update_seconds": 0.9577283732003338,
+    "producer_batches_per_second": 8.517641649529981,
+    "gpu_batches_per_second": 1.0441373859045355,
+    "producer_headroom": 8.157587080507756,
     "sustainable": true,
-    "buffer_seconds": 60.302662150504425,
+    "buffer_seconds": 60.33688751162103,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 321190.70556339057,
-    "upload_bytes_per_second": 5006058.920604437,
+    "generation_bytes_per_second": 356293.4437006376,
+    "upload_bytes_per_second": 5055778.385738575,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.92004847694672,
+    "estimated_full_shards_per_hour": 76.45227893127772,
     "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": 0.9817484880004486,
-      "bytes": 2589304,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.2340881659984007,
       "bytes": 16290946,
@@ -562,6 +556,12 @@
       "bytes": 11823249,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.1533994220008026,
+      "bytes": 7669461,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -240,3 +240,4 @@
 {"step": 421100, "loss": 0.5536261796951294, "lr": 0.0003, "grad_norm": 0.1453920134542661, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61637658452, "tokens_per_second": 151558.5535397857, "data_wait_s": 1.1459000234026462e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 421200, "loss": 0.5926738381385803, "lr": 0.0003, "grad_norm": 0.18534939670965112, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61652404313, "tokens_per_second": 152882.80334982387, "data_wait_s": 9.011000656755641e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 421300, "loss": 0.620008111000061, "lr": 0.0003, "grad_norm": 0.14913814283543417, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61667111139, "tokens_per_second": 152454.88758865822, "data_wait_s": 9.59899989538826e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 421400, "loss": 0.5742019414901733, "lr": 0.0003, "grad_norm": 0.13666841141816136, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61681766690, "tokens_per_second": 151368.23743257736, "data_wait_s": 1.0632000339683145e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:215ce636bc47e6913acb12cc1f27257238623985dac82134acf858de379c13d6
-size 516636
+oid sha256:e2141ebb99c3160a19509419aa04ccbfe54c6feeb8d2b9ba37226dab6d8ceeda
+size 518754