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": 74776.88341817356,
-      "required_pages_per_second_upper_estimate": 0.0064814707344089395
+      "required_token_positions_per_second": 74651.4059866145,
+      "required_pages_per_second_upper_estimate": 0.006470594668661043
     },
     {
       "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": 9347.110427271695,
-      "required_pages_per_second_upper_estimate": 0.0037287021011934316
+      "required_token_positions_per_second": 9331.425748326812,
+      "required_pages_per_second_upper_estimate": 0.00372244524825547
     },
     {
       "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": 18694.22085454339,
-      "required_pages_per_second_upper_estimate": 0.004324578987648559
+      "required_token_positions_per_second": 18662.851496653624,
+      "required_pages_per_second_upper_estimate": 0.004317322238782605
     },
     {
       "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": 28041.331281815084,
-      "required_pages_per_second_upper_estimate": 0.0017075699535674513
+      "required_token_positions_per_second": 27994.277244980432,
+      "required_pages_per_second_upper_estimate": 0.0017047046095976727
     },
     {
       "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": 28041.331281815084,
-      "required_pages_per_second_upper_estimate": 0.00215869930544723
+      "required_token_positions_per_second": 27994.277244980432,
+      "required_pages_per_second_upper_estimate": 0.002155076955437788
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28041.331281815084,
-      "required_pages_per_second_upper_estimate": 0.34817948622142725
+      "required_token_positions_per_second": 27994.277244980432,
+      "required_pages_per_second_upper_estimate": 0.34759523256367175
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.4155778188281545,
-    "producer_p95_seconds": 0.3382514959994296,
-    "gpu_update_seconds": 0.9859731594815485,
-    "producer_batches_per_second": 2.406288196082741,
-    "gpu_batches_per_second": 1.0142263918480572,
-    "producer_headroom": 2.3725355753148563,
+    "producer_batch_seconds": 0.10362380844543395,
+    "producer_p95_seconds": 0.3408884290001879,
+    "gpu_update_seconds": 0.9876304273923513,
+    "producer_batches_per_second": 9.650291906869823,
+    "gpu_batches_per_second": 1.0125244952611556,
+    "producer_headroom": 9.530921920442792,
     "sustainable": true,
-    "buffer_seconds": 60.14436272837446,
+    "buffer_seconds": 60.24545607093343,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 341109.2099537289,
-    "upload_bytes_per_second": 4007982.342240348,
+    "generation_bytes_per_second": 359104.0872754779,
+    "upload_bytes_per_second": 4003547.225777325,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.19409584006215,
+    "estimated_full_shards_per_hour": 77.05537761400464,
     "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": 1.1834909390017856,
-      "bytes": 7846092,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.0663465249999717,
       "bytes": 5706618,
@@ -562,6 +556,12 @@
       "bytes": 13283238,
       "pages": 34,
       "shards": 4
+    },
+    {
+      "seconds": 0.8937192140001571,
+      "bytes": 6207827,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -238,3 +238,4 @@
 {"step": 138800, "loss": 0.5648066718131304, "lr": 0.0003, "grad_norm": 0.15375385803934544, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20313041480, "tokens_per_second": 150122.57332411828, "data_wait_s": 1.0962998203467578e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 138900, "loss": 0.5808432884514332, "lr": 0.0003, "grad_norm": 0.15437790619225644, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20327720677, "tokens_per_second": 149687.26270518242, "data_wait_s": 1.1454001651145518e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 139000, "loss": 0.589487548917532, "lr": 0.0003, "grad_norm": 0.1428047880376676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20342441712, "tokens_per_second": 150873.92026288502, "data_wait_s": 1.2471999070839956e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 139100, "loss": 0.574429651722312, "lr": 0.0003, "grad_norm": 0.1376758639669584, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20357154219, "tokens_per_second": 150314.10776939144, "data_wait_s": 7.683000148972496e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:abe6a636d4dd6aa6200f33fbe5d5c52db4b06fd86c1487463fe452363e9a28bd
-size 511617
+oid sha256:7dd07251977f947216f5e76a6bbaeace8cf0ad7030f301be944cae5623d2b637
+size 513735