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": 76849.69019857602,
-      "required_pages_per_second_upper_estimate": 0.005316639883785879
+      "required_token_positions_per_second": 76826.66336006178,
+      "required_pages_per_second_upper_estimate": 0.005315046833667837
     },
     {
       "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": 9606.211274822002,
-      "required_pages_per_second_upper_estimate": 0.0040219200602655105
+      "required_token_positions_per_second": 9603.332920007722,
+      "required_pages_per_second_upper_estimate": 0.004020714953211655
     },
     {
       "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": 19212.422549644005,
-      "required_pages_per_second_upper_estimate": 0.004001440951436423
+      "required_token_positions_per_second": 19206.665840015445,
+      "required_pages_per_second_upper_estimate": 0.004000241980635436
     },
     {
       "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": 28818.633824466007,
-      "required_pages_per_second_upper_estimate": 0.0017276330942707727
+      "required_token_positions_per_second": 28809.998760023165,
+      "required_pages_per_second_upper_estimate": 0.0017271154353424045
     },
     {
       "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": 28818.633824466007,
-      "required_pages_per_second_upper_estimate": 0.002240522020078424
+      "required_token_positions_per_second": 28809.998760023165,
+      "required_pages_per_second_upper_estimate": 0.0022398506817996274
     },
     {
       "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": 28818.633824466007,
-      "required_pages_per_second_upper_estimate": 0.3696544916620619
+      "required_token_positions_per_second": 28809.998760023165,
+      "required_pages_per_second_upper_estimate": 0.36954373032699894
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07922623969528786,
-    "producer_p95_seconds": 0.1880708329999834,
-    "gpu_update_seconds": 0.9593792741322741,
-    "producer_batches_per_second": 12.622080813706434,
-    "gpu_batches_per_second": 1.0423406331187068,
-    "producer_headroom": 12.109362729092583,
+    "producer_batch_seconds": 0.0721565476406596,
+    "producer_p95_seconds": 0.18495447399982368,
+    "gpu_update_seconds": 0.9596668236711083,
+    "producer_batches_per_second": 13.85875617248224,
+    "gpu_batches_per_second": 1.0420283116327822,
+    "producer_headroom": 13.299788516078399,
     "sustainable": true,
-    "buffer_seconds": 60.44089427033327,
+    "buffer_seconds": 60.45900989127983,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 352873.13621839666,
-    "upload_bytes_per_second": 4857987.265565926,
+    "generation_bytes_per_second": 358842.8810371544,
+    "upload_bytes_per_second": 4855957.781070906,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.71836056627201,
+    "estimated_full_shards_per_hour": 76.99932883583044,
     "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": [
@@ -544,6 +544,12 @@
       "bytes": 7156662,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.3278628770003706,
+      "bytes": 6286061,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -72,3 +72,4 @@
 {"step": 404300, "loss": 0.5971654653549194, "lr": 0.0003, "grad_norm": 0.15261377341042853, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59179156258, "tokens_per_second": 154304.80222531557, "data_wait_s": 1.5270000403688755e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 404400, "loss": 0.5785234570503235, "lr": 0.0003, "grad_norm": 0.15924016359875773, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59193720644, "tokens_per_second": 150942.61431131698, "data_wait_s": 8.276999324152712e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 404500, "loss": 0.6061413288116455, "lr": 0.0003, "grad_norm": 0.16521312196485585, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59208247702, "tokens_per_second": 150981.2675535483, "data_wait_s": 9.579999641573522e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 404600, "loss": 0.5431584119796753, "lr": 0.0003, "grad_norm": 0.15124628593235345, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59222939708, "tokens_per_second": 152906.70188366118, "data_wait_s": 9.297999895352405e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d805ee925b7692d250a13734c9dc3056a1c44cb4f0a3bbc3fe396fd66a788f2f
-size 157680
+oid sha256:2b27586c6ca48cac8d075dc8a5f8e1f1cc71fe26f74383e6194b4befeda330c9
+size 159798