Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76030.12701058196,
-      "required_pages_per_second_upper_estimate": 0.006590098712682054
+      "required_token_positions_per_second": 75939.0768193936,
+      "required_pages_per_second_upper_estimate": 0.006582206712874446
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9503.765876322745,
-      "required_pages_per_second_upper_estimate": 0.003791194302027583
+      "required_token_positions_per_second": 9492.3846024242,
+      "required_pages_per_second_upper_estimate": 0.0037866541417042442
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19007.53175264549,
-      "required_pages_per_second_upper_estimate": 0.004397057949841003
+      "required_token_positions_per_second": 18984.7692048484,
+      "required_pages_per_second_upper_estimate": 0.004391792235015304
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28511.297628968237,
-      "required_pages_per_second_upper_estimate": 0.0017361884383862158
+      "required_token_positions_per_second": 28477.153807272596,
+      "required_pages_per_second_upper_estimate": 0.0017341092587837392
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28511.297628968237,
-      "required_pages_per_second_upper_estimate": 0.0021948786158011934
+      "required_token_positions_per_second": 28477.153807272596,
+      "required_pages_per_second_upper_estimate": 0.002192250129890915
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28511.297628968237,
-      "required_pages_per_second_upper_estimate": 0.35401489537688563
+      "required_token_positions_per_second": 28477.153807272596,
+      "required_pages_per_second_upper_estimate": 0.35359094338344604
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.09899706028116384,
-    "producer_p95_seconds": 0.33419607800169615,
-    "gpu_update_seconds": 0.9697208580190646,
-    "producer_batches_per_second": 10.101310050620462,
-    "gpu_batches_per_second": 1.031224595955159,
-    "producer_headroom": 9.795451049404274,
+    "producer_batch_seconds": 0.11826667167957794,
+    "producer_p95_seconds": 0.33940272399922833,
+    "gpu_update_seconds": 0.9708835435983478,
+    "producer_batches_per_second": 8.455467510824336,
+    "gpu_batches_per_second": 1.0299896487005424,
+    "producer_headroom": 8.209274259689835,
     "sustainable": true,
-    "buffer_seconds": 60.122693197182,
+    "buffer_seconds": 60.194779703097566,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 467075.74013241916,
-    "upload_bytes_per_second": 3528978.2493145126,
+    "generation_bytes_per_second": 487267.07296152326,
+    "upload_bytes_per_second": 3508235.566018321,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 100.22358086566382,
+    "estimated_full_shards_per_hour": 104.55617086061739,
     "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": 1.4853517330011528,
-      "bytes": 11092996,
-      "pages": 24,
-      "shards": 4
-    },
     {
       "seconds": 0.8748619340003643,
       "bytes": 6384371,
@@ -562,6 +556,12 @@
       "bytes": 5974861,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.2107169869996142,
+      "bytes": 7639876,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -257,3 +257,4 @@
 {"step": 65700, "loss": 0.8951286841183901, "lr": 0.0003, "grad_norm": 0.1622528794475597, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9615501451, "tokens_per_second": 150096.3029093902, "data_wait_s": 9.61700061452575e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 65800, "loss": 0.6213799938559532, "lr": 0.0003, "grad_norm": 0.13563732939225456, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9630064669, "tokens_per_second": 148455.73038975097, "data_wait_s": 1.2536998838186264e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 65900, "loss": 0.6381329819560051, "lr": 0.0003, "grad_norm": 0.2403724516363505, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9644776326, "tokens_per_second": 150310.7531065863, "data_wait_s": 1.0716998076532036e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 66000, "loss": 0.6565835401415825, "lr": 0.0003, "grad_norm": 0.1419560230238423, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9659487014, "tokens_per_second": 149725.03250765483, "data_wait_s": 1.1422998795751482e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c06f4b0735a9c907977100496fbaf185a4db90ee3385acfd005ff0705fbab89d
-size 552642
+oid sha256:1e9f13e54ba5fdbf9580a53ed41646d8b28af57559adbb7f866f3aebed150545
+size 554760