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": 76275.02857305737,
-      "required_pages_per_second_upper_estimate": 0.005276883459133821
+      "required_token_positions_per_second": 76497.9027306762,
+      "required_pages_per_second_upper_estimate": 0.005292302410497185
     },
     {
       "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": 9534.37857163217,
-      "required_pages_per_second_upper_estimate": 0.003991845207477346
+      "required_token_positions_per_second": 9562.237841334525,
+      "required_pages_per_second_upper_estimate": 0.004003509301934015
     },
     {
       "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": 19068.75714326434,
-      "required_pages_per_second_upper_estimate": 0.003971519235999085
+      "required_token_positions_per_second": 19124.47568266905,
+      "required_pages_per_second_upper_estimate": 0.003983123938360388
     },
     {
       "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": 28603.13571489651,
-      "required_pages_per_second_upper_estimate": 0.0017147143116486424
+      "required_token_positions_per_second": 28686.71352400358,
+      "required_pages_per_second_upper_estimate": 0.0017197246736886857
     },
     {
       "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": 28603.13571489651,
-      "required_pages_per_second_upper_estimate": 0.002223767989935405
+      "required_token_positions_per_second": 28686.71352400358,
+      "required_pages_per_second_upper_estimate": 0.0022302657969735487
     },
     {
       "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": 28603.13571489651,
-      "required_pages_per_second_upper_estimate": 0.366890313296347
+      "required_token_positions_per_second": 28686.71352400358,
+      "required_pages_per_second_upper_estimate": 0.36796235969271274
     }
   ],
   "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": 0.09226781845310938,
+    "producer_batch_seconds": 0.2339715080545659,
     "producer_p95_seconds": 0.27754543499759166,
-    "gpu_update_seconds": 0.9666073075198158,
-    "producer_batches_per_second": 10.83801499553391,
-    "gpu_batches_per_second": 1.0345462859843935,
-    "producer_headroom": 10.476104493692421,
+    "gpu_update_seconds": 0.9637911284910893,
+    "producer_batches_per_second": 4.2740246806751525,
+    "gpu_batches_per_second": 1.0375692102142497,
+    "producer_headroom": 4.119267070186673,
     "sustainable": true,
-    "buffer_seconds": 60.8962603737484,
+    "buffer_seconds": 60.718841094938625,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 167,
+    "upload_pages_per_commit": 158,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 636033.0404543991,
-    "upload_bytes_per_second": 4913846.067093352,
+    "generation_bytes_per_second": 601131.2274457037,
+    "upload_bytes_per_second": 4909252.917089508,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.238942195053,
+    "estimated_full_shards_per_hour": 64.4943838955323,
     "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": 1.2074203040010616,
-      "bytes": 7463350,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.190478869000799,
       "bytes": 1226986,
@@ -562,6 +556,12 @@
       "bytes": 5756694,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.087463227002445,
+      "bytes": 6477149,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -215,3 +215,4 @@
 {"step": 906500, "loss": 0.5194767713546753, "lr": 3e-05, "grad_norm": 0.319990316796002, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132685900827, "tokens_per_second": 152826.49416012273, "data_wait_s": 1.9835002603940666e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 906600, "loss": 0.527674674987793, "lr": 3e-05, "grad_norm": 0.34377388706849443, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132700371678, "tokens_per_second": 150210.87395500537, "data_wait_s": 1.0879000910790637e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 906700, "loss": 0.5318596363067627, "lr": 3e-05, "grad_norm": 0.3319238812255569, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132714852166, "tokens_per_second": 150475.67974385415, "data_wait_s": 1.4097000530455261e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 906800, "loss": 0.5484407544136047, "lr": 3e-05, "grad_norm": 0.3447702822185023, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132729456073, "tokens_per_second": 151491.73993199418, "data_wait_s": 1.410899858456105e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4ca7c7ba62bac8da894ac9abd2ecc0251f1b5868213e0c4abc9bc1f62c112e0c
-size 462903
+oid sha256:f4ab5a55b0eb6d38aa2f99ad021d4b88f9018ffecf15dd029cde1488595ad556
+size 465021