Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75486.33652598412,
-      "required_pages_per_second_upper_estimate": 0.006542964331701496
+      "required_token_positions_per_second": 75198.37537617187,
+      "required_pages_per_second_upper_estimate": 0.006518004589066631
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9435.792065748015,
-      "required_pages_per_second_upper_estimate": 0.0037640785326902887
+      "required_token_positions_per_second": 9399.796922021484,
+      "required_pages_per_second_upper_estimate": 0.003749719531682417
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18871.58413149603,
-      "required_pages_per_second_upper_estimate": 0.00436560886028449
+      "required_token_positions_per_second": 18799.593844042967,
+      "required_pages_per_second_upper_estimate": 0.0043489551742680104
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28307.376197244044,
-      "required_pages_per_second_upper_estimate": 0.001723770693087279
+      "required_token_positions_per_second": 28199.390766064455,
+      "required_pages_per_second_upper_estimate": 0.0017171949468842136
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28307.376197244044,
-      "required_pages_per_second_upper_estimate": 0.002179180179496413
+      "required_token_positions_per_second": 28199.390766064455,
+      "required_pages_per_second_upper_estimate": 0.0021708671620814025
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28307.376197244044,
-      "required_pages_per_second_upper_estimate": 0.3514828736760004
+      "required_token_positions_per_second": 28199.390766064455,
+      "required_pages_per_second_upper_estimate": 0.35014205602473963
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.0885385873673954,
-    "producer_p95_seconds": 0.27832846700039227,
-    "gpu_update_seconds": 0.9767065590025175,
-    "producer_batches_per_second": 11.294510447184445,
-    "gpu_batches_per_second": 1.0238489654674496,
-    "producer_headroom": 11.031422434487505,
+    "producer_batch_seconds": 0.08078218754698696,
+    "producer_p95_seconds": 0.2882196380014648,
+    "gpu_update_seconds": 0.9804467135251729,
+    "producer_batches_per_second": 12.378966581195266,
+    "gpu_batches_per_second": 1.019943242406845,
+    "producer_headroom": 12.136917101370845,
     "sustainable": true,
-    "buffer_seconds": 60.55580665815608,
+    "buffer_seconds": 60.78769623856072,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 245492.2625791077,
-    "upload_bytes_per_second": 5260478.298811827,
+    "generation_bytes_per_second": 243909.70507795224,
+    "upload_bytes_per_second": 5316427.706025327,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 52.67692478208468,
+    "estimated_full_shards_per_hour": 52.337344782389884,
     "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": 1.0232497649994912,
-      "bytes": 1108747,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.292045906000567,
       "bytes": 13335425,
@@ -562,6 +556,12 @@
       "bytes": 8287753,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.1439757010011817,
+      "bytes": 6251248,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -241,3 +241,4 @@
 {"step": 379100, "loss": 0.5877036396414042, "lr": 0.0003, "grad_norm": 0.20884160123560638, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55488828045, "tokens_per_second": 146873.64116449817, "data_wait_s": 1.1002000974258408e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 379200, "loss": 0.5972409769892693, "lr": 0.0003, "grad_norm": 0.16443295568660182, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55503586691, "tokens_per_second": 151127.7999481926, "data_wait_s": 1.701300061540678e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 379300, "loss": 0.5972128799185157, "lr": 0.0003, "grad_norm": 0.16542582131713449, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55518278840, "tokens_per_second": 149661.1695557182, "data_wait_s": 9.7760021162685e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 379400, "loss": 0.5842984765768051, "lr": 0.0003, "grad_norm": 0.1710354123652569, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55532897833, "tokens_per_second": 149429.39549429552, "data_wait_s": 1.1393000022508204e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:44e34f9dae5d82142ec4518a88aee91b47df661a54b8fb6e6106b68027f69a08
-size 517971
+oid sha256:2d1873d1ab1fac7be94532756d1242aca896784be93595ff5b917ab27bd872a7
+size 520089