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": 76165.47545599737,
-      "required_pages_per_second_upper_estimate": 0.005269304320297357
+      "required_token_positions_per_second": 75390.81684050769,
+      "required_pages_per_second_upper_estimate": 0.005215711639822152
     },
     {
       "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": 9520.68443199967,
-      "required_pages_per_second_upper_estimate": 0.003986111757179372
+      "required_token_positions_per_second": 9423.85210506346,
+      "required_pages_per_second_upper_estimate": 0.0039455700839801065
     },
     {
       "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": 19041.36886399934,
-      "required_pages_per_second_upper_estimate": 0.003965814979705681
+      "required_token_positions_per_second": 18847.70421012692,
+      "required_pages_per_second_upper_estimate": 0.0039254797395844275
     },
     {
       "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": 28562.053295999012,
-      "required_pages_per_second_upper_estimate": 0.0017122514833649589
+      "required_token_positions_per_second": 28271.556315190384,
+      "required_pages_per_second_upper_estimate": 0.001694836632928688
     },
     {
       "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": 28562.053295999012,
-      "required_pages_per_second_upper_estimate": 0.002220574012568588
+      "required_token_positions_per_second": 28271.556315190384,
+      "required_pages_per_second_upper_estimate": 0.002197989150071898
     },
     {
       "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": 28562.053295999012,
-      "required_pages_per_second_upper_estimate": 0.3663633521375946
+      "required_token_positions_per_second": 28271.556315190384,
+      "required_pages_per_second_upper_estimate": 0.3626371687791381
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07952877826562599,
-    "producer_p95_seconds": 0.2313084589986829,
-    "gpu_update_seconds": 0.9679976335550409,
-    "producer_batches_per_second": 12.57406465694722,
-    "gpu_batches_per_second": 1.0330603767360753,
-    "producer_headroom": 12.171664832092986,
+    "producer_batch_seconds": 0.09489289803897805,
+    "producer_p95_seconds": 0.2201381439990655,
+    "gpu_update_seconds": 0.9779440400012451,
+    "producer_batches_per_second": 10.538196436884473,
+    "gpu_batches_per_second": 1.022553396816782,
+    "producer_headroom": 10.305766397813528,
     "sustainable": true,
-    "buffer_seconds": 60.015853280412536,
+    "buffer_seconds": 60.6325304800772,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 470968.6321024544,
-    "upload_bytes_per_second": 5719273.560609667,
+    "generation_bytes_per_second": 468501.08164653654,
+    "upload_bytes_per_second": 5652535.764722333,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.05890486054633,
+    "estimated_full_shards_per_hour": 100.52942597434112,
     "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": 1.502407841000604,
-      "bytes": 9864729,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 1.077067022999472,
       "bytes": 1093074,
@@ -562,6 +556,12 @@
       "bytes": 5981571,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 0.8534065610001562,
+      "bytes": 1138564,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -187,3 +187,4 @@
 {"step": 703700, "loss": 0.5930591225624084, "lr": 8e-05, "grad_norm": 0.22176475071326485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103002235850, "tokens_per_second": 153173.04245013505, "data_wait_s": 1.5246998373186216e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 703800, "loss": 0.5478715300559998, "lr": 8e-05, "grad_norm": 0.21695522373717682, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103016893084, "tokens_per_second": 152664.82882184125, "data_wait_s": 9.547999070491642e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 703900, "loss": 0.5815778970718384, "lr": 8e-05, "grad_norm": 0.22463245838636323, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103031557959, "tokens_per_second": 153466.13672966024, "data_wait_s": 9.208997653331608e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 704000, "loss": 0.5550633072853088, "lr": 8e-05, "grad_norm": 0.21852078942307152, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103046306084, "tokens_per_second": 152962.1840814142, "data_wait_s": 1.9200997485313565e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a58b6a25033db303b64743bab9d20c35cd319a0a88d6022721227504e3ec6fd5
-size 402816
+oid sha256:3db4b1a8efea6c1a6845bee5633ee71760cde6227a84e058cb30d6a36cbda802
+size 404934