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": 77170.13682853756,
-      "required_pages_per_second_upper_estimate": 0.0053388091251852406
+      "required_token_positions_per_second": 77125.44925162838,
+      "required_pages_per_second_upper_estimate": 0.005335717534925199
     },
     {
       "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": 9646.267103567196,
-      "required_pages_per_second_upper_estimate": 0.004038690599300301
+      "required_token_positions_per_second": 9640.681156453547,
+      "required_pages_per_second_upper_estimate": 0.0040363518798916575
     },
     {
       "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": 19292.53420713439,
-      "required_pages_per_second_upper_estimate": 0.00401812609700519
+      "required_token_positions_per_second": 19281.362312907095,
+      "required_pages_per_second_upper_estimate": 0.004015799286060308
     },
     {
       "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": 28938.801310701583,
-      "required_pages_per_second_upper_estimate": 0.0017348369515854682
+      "required_token_positions_per_second": 28922.043469360644,
+      "required_pages_per_second_upper_estimate": 0.001733832344584821
     },
     {
       "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": 28938.801310701583,
-      "required_pages_per_second_upper_estimate": 0.002249864513572328
+      "required_token_positions_per_second": 28922.043469360644,
+      "required_pages_per_second_upper_estimate": 0.0022485616651180245
     },
     {
       "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": 28938.801310701583,
-      "required_pages_per_second_upper_estimate": 0.3711958711496977
+      "required_token_positions_per_second": 28922.043469360644,
+      "required_pages_per_second_upper_estimate": 0.3709809195541443
     }
   ],
   "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": 0.08906326558602018,
-    "producer_p95_seconds": 0.25963005800076644,
-    "gpu_update_seconds": 0.955395481075982,
-    "producer_batches_per_second": 11.227973659175653,
-    "gpu_batches_per_second": 1.0466869687030376,
-    "producer_headroom": 10.727155295616578,
+    "producer_batch_seconds": 0.34113621535158245,
+    "producer_p95_seconds": 0.268373019000137,
+    "gpu_update_seconds": 0.9559490507401271,
+    "producer_batches_per_second": 2.931380354822129,
+    "gpu_batches_per_second": 1.0460808546499074,
+    "producer_headroom": 2.8022502675504715,
     "sustainable": true,
-    "buffer_seconds": 60.18991530778687,
+    "buffer_seconds": 60.22479019662801,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 358072.56606147625,
+    "generation_bytes_per_second": 331034.67553161556,
     "upload_bytes_per_second": 3846490.20823898,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.83403717406479,
+    "estimated_full_shards_per_hour": 71.0323352762351,
     "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": [

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -273,3 +273,4 @@
 {"step": 552300, "loss": 0.5912648439407349, "lr": 0.0003, "grad_norm": 0.14267824863647247, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80839119291, "tokens_per_second": 150291.8241835844, "data_wait_s": 1.692500154604204e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 552400, "loss": 0.5595518350601196, "lr": 0.0003, "grad_norm": 0.1525354226985604, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80853665610, "tokens_per_second": 151328.334717717, "data_wait_s": 8.303999493364245e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 552500, "loss": 0.6008196473121643, "lr": 0.0003, "grad_norm": 0.16624135242016908, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80868153498, "tokens_per_second": 150446.95472104294, "data_wait_s": 1.3756998669123277e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 552600, "loss": 0.6145492196083069, "lr": 0.0003, "grad_norm": 0.16738370911393524, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80882804878, "tokens_per_second": 152742.51443957747, "data_wait_s": 1.19440010166727e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9dde805e1902adf0ac94613321a962e7d4fe7c12b3d338425183928e2536740c
-size 586530
+oid sha256:76faf5d5b6a3e3c233c6dd80fdfb0c45b06fdd4c98bfe436df2e138ac1682869
+size 588648