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": 75773.78078331477,
-      "required_pages_per_second_upper_estimate": 0.006567879271406106
+      "required_token_positions_per_second": 75679.87601902093,
+      "required_pages_per_second_upper_estimate": 0.006559739844436566
     },
     {
       "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": 9471.722597914346,
-      "required_pages_per_second_upper_estimate": 0.0037784117591807666
+      "required_token_positions_per_second": 9459.984502377616,
+      "required_pages_per_second_upper_estimate": 0.0037737292573709973
     },
     {
       "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": 18943.44519582869,
-      "required_pages_per_second_upper_estimate": 0.004382232652806315
+      "required_token_positions_per_second": 18919.969004755232,
+      "required_pages_per_second_upper_estimate": 0.004376801849168321
     },
     {
       "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": 28415.167793743036,
-      "required_pages_per_second_upper_estimate": 0.001730334635775268
+      "required_token_positions_per_second": 28379.95350713285,
+      "required_pages_per_second_upper_estimate": 0.0017281902704757907
     },
     {
       "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": 28415.167793743036,
-      "required_pages_per_second_upper_estimate": 0.002187478274981142
+      "required_token_positions_per_second": 28379.95350713285,
+      "required_pages_per_second_upper_estimate": 0.0021847673817185055
     },
     {
       "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": 28415.167793743036,
-      "required_pages_per_second_upper_estimate": 0.3528212845492511
+      "required_token_positions_per_second": 28379.95350713285,
+      "required_pages_per_second_upper_estimate": 0.3523840409641885
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.12474190689843212,
-    "producer_p95_seconds": 0.1442642329998307,
-    "gpu_update_seconds": 0.9730014688172292,
-    "producer_batches_per_second": 8.016552134434054,
-    "gpu_batches_per_second": 1.027747677725081,
-    "producer_headroom": 7.800117001654227,
+    "producer_batch_seconds": 0.04573885890625817,
+    "producer_p95_seconds": 0.1495671850002509,
+    "gpu_update_seconds": 0.9742087841352917,
+    "producer_batches_per_second": 21.863247661020594,
+    "gpu_batches_per_second": 1.0264740128447936,
+    "producer_headroom": 21.299367921091633,
     "sustainable": true,
-    "buffer_seconds": 60.326091066668205,
+    "buffer_seconds": 60.40094461638808,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 148,
+    "upload_pages_per_commit": 176,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 638677.8404645902,
-    "upload_bytes_per_second": 2633196.0864612823,
+    "generation_bytes_per_second": 593174.6967376038,
+    "upload_bytes_per_second": 2734484.827849634,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.52269845224991,
+    "estimated_full_shards_per_hour": 63.64074075983088,
     "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": [
@@ -286,6 +286,12 @@
       "bytes": 2480897,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.156973391000065,
+      "bytes": 7369438,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -16,3 +16,4 @@
 {"step": 1600, "loss": 0.8880256526172161, "lr": 0.0003, "grad_norm": 0.22562145193875374, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 234611375, "tokens_per_second": 149071.75160437857, "data_wait_s": 8.28099973659846e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 1700, "loss": 0.9110855236649513, "lr": 0.0003, "grad_norm": 0.2608335000330014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 249356749, "tokens_per_second": 150226.706145175, "data_wait_s": 1.4890999864292098e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 1800, "loss": 0.9287391342222691, "lr": 0.0003, "grad_norm": 0.23552261001673377, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 263963539, "tokens_per_second": 149206.06424722384, "data_wait_s": 1.4887000361341052e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 1900, "loss": 0.9189868532121181, "lr": 0.0003, "grad_norm": 0.23279412071070146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 278648795, "tokens_per_second": 148790.8751210036, "data_wait_s": 1.0176000159844989e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:426833c87931dffc5afab24b0c5789a2f1670cf4c349e05621598e1a125824db
-size 37620
+oid sha256:12d072ea19129dc796df73170a653ad0a506c0eb8403b3d76a118549ea420556
+size 39703