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": 76719.65543141075,
-      "required_pages_per_second_upper_estimate": 0.005307643776871157
+      "required_token_positions_per_second": 76905.86179778133,
+      "required_pages_per_second_upper_estimate": 0.005320525965354969
     },
     {
       "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": 9589.956928926344,
-      "required_pages_per_second_upper_estimate": 0.004015114705068338
+      "required_token_positions_per_second": 9613.232724722666,
+      "required_pages_per_second_upper_estimate": 0.004024859794714371
     },
     {
       "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": 19179.913857852687,
-      "required_pages_per_second_upper_estimate": 0.003994670248248146
+      "required_token_positions_per_second": 19226.46544944533,
+      "required_pages_per_second_upper_estimate": 0.004004365717128867
     },
     {
       "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": 28769.870786779033,
-      "required_pages_per_second_upper_estimate": 0.001724709824618286
+      "required_token_positions_per_second": 28839.698174168,
+      "required_pages_per_second_upper_estimate": 0.0017288958698720076
     },
     {
       "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": 28769.870786779033,
-      "required_pages_per_second_upper_estimate": 0.002236730908384197
+      "required_token_positions_per_second": 28839.698174168,
+      "required_pages_per_second_upper_estimate": 0.0022421596806154716
     },
     {
       "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": 28769.870786779033,
-      "required_pages_per_second_upper_estimate": 0.369029011772284
+      "required_token_positions_per_second": 28839.698174168,
+      "required_pages_per_second_upper_estimate": 0.36992468252290245
     }
   ],
   "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": 0.06760630871103501,
-    "producer_p95_seconds": 0.22482100800152693,
-    "gpu_update_seconds": 0.9610053588667983,
-    "producer_batches_per_second": 14.791519002675196,
-    "gpu_batches_per_second": 1.0405769237116258,
-    "producer_headroom": 14.214729027350943,
+    "producer_batch_seconds": 0.10834756593770578,
+    "producer_p95_seconds": 0.21621506999872508,
+    "gpu_update_seconds": 0.9586785490274161,
+    "producer_batches_per_second": 9.22955667112031,
+    "gpu_batches_per_second": 1.0431025091929977,
+    "producer_headroom": 8.848177997635927,
     "sustainable": true,
-    "buffer_seconds": 60.54333760860829,
+    "buffer_seconds": 60.39674858872721,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 448066.3681804032,
-    "upload_bytes_per_second": 3445912.4400736447,
+    "generation_bytes_per_second": 432291.1183218623,
+    "upload_bytes_per_second": 3454333.1690802337,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.14461216029235,
+    "estimated_full_shards_per_hour": 92.75961076967145,
     "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": 1.541119751000224,
-      "bytes": 6183503,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.4027249899991148,
       "bytes": 1181995,
@@ -562,6 +556,12 @@
       "bytes": 9992623,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.2029927740004496,
+      "bytes": 6068424,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -163,3 +163,4 @@
 {"step": 858800, "loss": 0.498494952917099, "lr": 3e-05, "grad_norm": 0.29280652892263864, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125704548608, "tokens_per_second": 150410.5881663116, "data_wait_s": 1.1326999810989946e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 858900, "loss": 0.5148037075996399, "lr": 3e-05, "grad_norm": 0.3186130609277609, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125719270141, "tokens_per_second": 152950.19550174504, "data_wait_s": 1.1157999324495904e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 859000, "loss": 0.490352600812912, "lr": 3e-05, "grad_norm": 0.29580371604890127, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125733757520, "tokens_per_second": 150539.92195541682, "data_wait_s": 1.2584998330567032e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 859100, "loss": 0.5722727179527283, "lr": 3e-05, "grad_norm": 0.31890911253668647, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125748278219, "tokens_per_second": 150417.78949822113, "data_wait_s": 1.3102002412779257e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:10a90dc41db9f3b3eb2380813ec8de46aa6b7622326f6b911ae85afd4146e08a
-size 351984
+oid sha256:93581846b87aa14205e3147a1e8b4788198f2de7b38ebe635d2f13e58d4a926a
+size 354102