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": 77403.74675250174,
-      "required_pages_per_second_upper_estimate": 0.005354970801774785
+      "required_token_positions_per_second": 76408.23690692343,
+      "required_pages_per_second_upper_estimate": 0.005286099120756594
     },
     {
       "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": 9675.468344062718,
-      "required_pages_per_second_upper_estimate": 0.004050916548904534
+      "required_token_positions_per_second": 9551.029613365428,
+      "required_pages_per_second_upper_estimate": 0.0039988166509377695
     },
     {
       "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": 19350.936688125435,
-      "required_pages_per_second_upper_estimate": 0.0040302897936186285
+      "required_token_positions_per_second": 19102.059226730857,
+      "required_pages_per_second_upper_estimate": 0.00397845518175016
     },
     {
       "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": 29026.405032188155,
-      "required_pages_per_second_upper_estimate": 0.0017400886609254507
+      "required_token_positions_per_second": 28653.088840096283,
+      "required_pages_per_second_upper_estimate": 0.001717708925230363
     },
     {
       "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": 29026.405032188155,
-      "required_pages_per_second_upper_estimate": 0.002256675317589861
+      "required_token_positions_per_second": 28653.088840096283,
+      "required_pages_per_second_upper_estimate": 0.0022276516256991103
     },
     {
       "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": 29026.405032188155,
-      "required_pages_per_second_upper_estimate": 0.3723195576273798
+      "required_token_positions_per_second": 28653.088840096283,
+      "required_pages_per_second_upper_estimate": 0.3675310583509227
     }
   ],
   "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": 704566,
-    "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,
+    "producer_batch_seconds": 0.08638957732034669,
+    "producer_p95_seconds": 0.23644456900001387,
+    "gpu_update_seconds": 0.9649221469383157,
+    "producer_batches_per_second": 11.575470456254651,
+    "gpu_batches_per_second": 1.0363530396446863,
+    "producer_headroom": 11.169427804470283,
     "sustainable": true,
-    "buffer_seconds": 60.00825793164689,
+    "buffer_seconds": 60.79009525711389,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 337189.5586334336,
-    "upload_bytes_per_second": 5400427.456766236,
+    "generation_bytes_per_second": 304275.39319865545,
+    "upload_bytes_per_second": 5376519.478281569,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.35302991153961,
+    "estimated_full_shards_per_hour": 65.2904162117934,
     "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": 1.9781427679999979,
-      "bytes": 8201204,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 0.9253323299999465,
       "bytes": 6468324,
@@ -562,6 +556,12 @@
       "bytes": 259525,
       "pages": 8,
       "shards": 1
+    },
+    {
+      "seconds": 0.9178394819991809,
+      "bytes": 643385,
+      "pages": 18,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -129,3 +129,4 @@
 {"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}
+{"step": 410300, "loss": 0.5704032182693481, "lr": 0.0003, "grad_norm": 0.16303334227756025, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60057065977, "tokens_per_second": 152063.24380120283, "data_wait_s": 1.6596999557805248e-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:ba40389223008195704f3fde97aa1ff50d1b00d5472f702b04a5128277eff935
-size 279972
+oid sha256:f3f8982dacbaa657e5a5ba703fbeacbdb7d54ac226aff0c23b6879550732ade0
+size 282090