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": 76487.28368217818,
-      "required_pages_per_second_upper_estimate": 0.005291567760082505
+      "required_token_positions_per_second": 77403.74675250174,
+      "required_pages_per_second_upper_estimate": 0.005354970801774785
     },
     {
       "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": 9560.910460272273,
-      "required_pages_per_second_upper_estimate": 0.004002953555202119
+      "required_token_positions_per_second": 9675.468344062718,
+      "required_pages_per_second_upper_estimate": 0.004050916548904534
     },
     {
       "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": 19121.820920544546,
-      "required_pages_per_second_upper_estimate": 0.003982571021420639
+      "required_token_positions_per_second": 19350.936688125435,
+      "required_pages_per_second_upper_estimate": 0.0040302897936186285
     },
     {
       "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": 28682.73138081682,
-      "required_pages_per_second_upper_estimate": 0.0017194859502850201
+      "required_token_positions_per_second": 29026.405032188155,
+      "required_pages_per_second_upper_estimate": 0.0017400886609254507
     },
     {
       "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": 28682.73138081682,
-      "required_pages_per_second_upper_estimate": 0.002229956202856375
+      "required_token_positions_per_second": 29026.405032188155,
+      "required_pages_per_second_upper_estimate": 0.002256675317589861
     },
     {
       "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": 28682.73138081682,
-      "required_pages_per_second_upper_estimate": 0.3679112810356052
+      "required_token_positions_per_second": 29026.405032188155,
+      "required_pages_per_second_upper_estimate": 0.3723195576273798
     }
   ],
   "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": 0.06967086632812425,
-    "producer_p95_seconds": 0.2758707979992323,
-    "gpu_update_seconds": 0.9639249356318675,
-    "producer_batches_per_second": 14.3532017427538,
-    "gpu_batches_per_second": 1.0374251801510712,
-    "producer_headroom": 13.835409065995165,
+    "producer_batch_seconds": 0.06591298851566307,
+    "producer_p95_seconds": 0.25100682500124094,
+    "gpu_update_seconds": 0.9525120306610617,
+    "producer_batches_per_second": 15.171516608784435,
+    "gpu_batches_per_second": 1.0498555060831942,
+    "producer_headroom": 14.451052093241286,
     "sustainable": true,
-    "buffer_seconds": 60.72727094480766,
+    "buffer_seconds": 60.00825793164689,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 338055.88337438414,
+    "generation_bytes_per_second": 337189.5586334336,
     "upload_bytes_per_second": 5400427.456766236,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.53892303393977,
+    "estimated_full_shards_per_hour": 72.35302991153961,
     "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-v1/1790824801708454555-41033672/train.jsonl

@@ -128,3 +128,4 @@
 {"step": 409900, "loss": 0.5718177556991577, "lr": 0.0003, "grad_norm": 0.139769420556045, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59998617478, "tokens_per_second": 152259.28878810006, "data_wait_s": 9.601999408914708e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 410000, "loss": 0.5736613273620605, "lr": 0.0003, "grad_norm": 0.18205656433172238, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60013390275, "tokens_per_second": 153248.6850233319, "data_wait_s": 8.675999197294004e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 410100, "loss": 0.609076201915741, "lr": 0.0003, "grad_norm": 0.15870267522321527, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60027944595, "tokens_per_second": 140703.30694923893, "data_wait_s": 1.1575999451451935e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 410200, "loss": 0.5726893544197083, "lr": 0.0003, "grad_norm": 0.15037622728187877, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60042474665, "tokens_per_second": 125187.33204144836, "data_wait_s": 1.0036999810836278e-05, "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 sha256:25990c67ed7908cac301544d63f01e2764b7a459058f70d5e13325a352c8cec8
-size 277854
+oid sha256:ba40389223008195704f3fde97aa1ff50d1b00d5472f702b04a5128277eff935
+size 279972