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": 75289.349934471,
-      "required_pages_per_second_upper_estimate": 0.00652589003320709
+      "required_token_positions_per_second": 75434.92562306488,
+      "required_pages_per_second_upper_estimate": 0.006538508164936198
     },
     {
       "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": 9411.168741808875,
-      "required_pages_per_second_upper_estimate": 0.003754255920619465
+      "required_token_positions_per_second": 9429.36570288311,
+      "required_pages_per_second_upper_estimate": 0.0037615149604607904
     },
     {
       "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": 18822.33748361775,
-      "required_pages_per_second_upper_estimate": 0.004354216515009251
+      "required_token_positions_per_second": 18858.73140576622,
+      "required_pages_per_second_upper_estimate": 0.0043626356083871455
     },
     {
       "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": 28233.506225426623,
-      "required_pages_per_second_upper_estimate": 0.0017192723993693876
+      "required_token_positions_per_second": 28288.097108649326,
+      "required_pages_per_second_upper_estimate": 0.0017225966977414213
     },
     {
       "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": 28233.506225426623,
-      "required_pages_per_second_upper_estimate": 0.0021734934645807383
+      "required_token_positions_per_second": 28288.097108649326,
+      "required_pages_per_second_upper_estimate": 0.0021776960218884587
     },
     {
       "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": 28233.506225426623,
-      "required_pages_per_second_upper_estimate": 0.35056565585292004
+      "required_token_positions_per_second": 28288.097108649326,
+      "required_pages_per_second_upper_estimate": 0.35124349191861287
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.49403402603888935,
-    "producer_p95_seconds": 0.36854817899438785,
-    "gpu_update_seconds": 0.9792620080286264,
-    "producer_batches_per_second": 2.024152077171466,
-    "gpu_batches_per_second": 1.0211771638247478,
-    "producer_headroom": 1.982175227646245,
+    "producer_batch_seconds": 0.10696796451537693,
+    "producer_p95_seconds": 0.35885942399909254,
+    "gpu_update_seconds": 0.9773722104321534,
+    "producer_batches_per_second": 9.348593333813016,
+    "gpu_batches_per_second": 1.0231516604690873,
+    "producer_headroom": 9.137055331100122,
     "sustainable": true,
-    "buffer_seconds": 60.714244497774835,
+    "buffer_seconds": 60.59707704679351,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 252987.79977760246,
-    "upload_bytes_per_second": 5723002.402717545,
+    "generation_bytes_per_second": 262317.8338768617,
+    "upload_bytes_per_second": 5764440.892628973,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.28529257770591,
+    "estimated_full_shards_per_hour": 56.28730070332898,
     "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": 0.8301882180021494,
-      "bytes": 1174116,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.4423049569995783,
       "bytes": 10949179,
@@ -562,6 +556,12 @@
       "bytes": 11991624,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 1.2643025360011961,
+      "bytes": 6743722,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -267,3 +267,4 @@
 {"step": 306900, "loss": 0.5960936769843102, "lr": 0.0003, "grad_norm": 0.15019100022239507, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44921851435, "tokens_per_second": 150143.078845326, "data_wait_s": 9.848001354839653e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 307000, "loss": 0.6174443885684013, "lr": 0.0003, "grad_norm": 0.23231451502623285, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44936485048, "tokens_per_second": 141820.6529797209, "data_wait_s": 1.0821997420862317e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 307100, "loss": 0.5371118113398552, "lr": 0.0003, "grad_norm": 0.19132379531914268, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44951135665, "tokens_per_second": 131276.77808792147, "data_wait_s": 1.3419004972092807e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 307200, "loss": 0.6113442908972502, "lr": 0.0003, "grad_norm": 0.14900561915782579, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44965784277, "tokens_per_second": 150229.627348448, "data_wait_s": 1.150499883806333e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1bf7a80b85dcdb49054a78f1f54a6992d0c53b86e17781c110f465cb52b71f5f
-size 573822
+oid sha256:f76099a9f333aa078f32400f8c2d5ee81fc43cccac5a437eecf8abf7039cb4a4
+size 575940