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": 77233.79034655003,
-      "required_pages_per_second_upper_estimate": 0.0053432128232578525
+      "required_token_positions_per_second": 76937.98874084065,
+      "required_pages_per_second_upper_estimate": 0.005322748581820591
     },
     {
       "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": 9654.223793318753,
-      "required_pages_per_second_upper_estimate": 0.004042021899144703
+      "required_token_positions_per_second": 9617.248592605081,
+      "required_pages_per_second_upper_estimate": 0.004026541154735881
     },
     {
       "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": 19308.447586637507,
-      "required_pages_per_second_upper_estimate": 0.004021440434291651
+      "required_token_positions_per_second": 19234.497185210163,
+      "required_pages_per_second_upper_estimate": 0.0040060385158775655
     },
     {
       "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": 28962.671379956264,
-      "required_pages_per_second_upper_estimate": 0.001736267925789801
+      "required_token_positions_per_second": 28851.745777815242,
+      "required_pages_per_second_upper_estimate": 0.001729618105315034
     },
     {
       "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": 28962.671379956264,
-      "required_pages_per_second_upper_estimate": 0.002251720306463541
+      "required_token_positions_per_second": 28851.745777815242,
+      "required_pages_per_second_upper_estimate": 0.0022430963288072366
     },
     {
       "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": 28962.671379956264,
-      "required_pages_per_second_upper_estimate": 0.37150205076841325
+      "required_token_positions_per_second": 28851.745777815242,
+      "required_pages_per_second_upper_estimate": 0.37007921624678036
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.13133296921870397,
-    "producer_p95_seconds": 0.37477072100591613,
-    "gpu_update_seconds": 0.9546080759364591,
-    "producer_batches_per_second": 7.614234308026165,
-    "gpu_batches_per_second": 1.0475503247958717,
-    "producer_headroom": 7.268609562514234,
+    "producer_batch_seconds": 0.1195175636638055,
+    "producer_p95_seconds": 0.3620480229947134,
+    "gpu_update_seconds": 0.9582782342848443,
+    "producer_batches_per_second": 8.366971090650154,
+    "gpu_batches_per_second": 1.043538258746211,
+    "producer_headroom": 8.017886283060568,
     "sustainable": true,
-    "buffer_seconds": 60.14030878399693,
+    "buffer_seconds": 60.371528759945186,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 322247.5395725463,
-    "upload_bytes_per_second": 5011841.984188898,
+    "generation_bytes_per_second": 347894.1334572505,
+    "upload_bytes_per_second": 4978089.031321231,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.14682045347493,
+    "estimated_full_shards_per_hour": 74.64998247898231,
     "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": 1.808516741002677,
-      "bytes": 10401411,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.1967187990012462,
       "bytes": 13416312,
@@ -562,6 +556,12 @@
       "bytes": 5817685,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.866520342999138,
+      "bytes": 7735960,
+      "pages": 41,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -420,3 +420,4 @@
 {"step": 687000, "loss": 0.5700095295906067, "lr": 0.0001, "grad_norm": 0.2111146027716485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100558815263, "tokens_per_second": 152244.84273307477, "data_wait_s": 1.2853997759521008e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 687100, "loss": 0.5151077508926392, "lr": 0.0001, "grad_norm": 0.20409389195283212, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100573505138, "tokens_per_second": 151924.68522414909, "data_wait_s": 1.2672004231717438e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 687200, "loss": 0.587519645690918, "lr": 0.0001, "grad_norm": 0.21587036717376892, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100588032378, "tokens_per_second": 150816.02977310724, "data_wait_s": 1.3817996659781784e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 687300, "loss": 0.5630563497543335, "lr": 0.0001, "grad_norm": 0.20577558342061913, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100602552012, "tokens_per_second": 150811.66896962764, "data_wait_s": 1.1127995094284415e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:af8a2c644da4f1105ed5710fa0f0bab7ed337534f98aa55525fedd4dce798462
-size 900225
+oid sha256:169677a926f2edcddd59f516c64bf1f72e24cbe84fff97a8e3a05ce449061ca8
+size 902343