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": 75508.24926500011,
-      "required_pages_per_second_upper_estimate": 0.0065448636723820725
+      "required_token_positions_per_second": 75867.60597007182,
+      "required_pages_per_second_upper_estimate": 0.006576011800796446
     },
     {
       "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": 9438.531158125013,
-      "required_pages_per_second_upper_estimate": 0.0037651711975925537
+      "required_token_positions_per_second": 9483.450746258977,
+      "required_pages_per_second_upper_estimate": 0.0037830902929068843
     },
     {
       "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": 18877.062316250027,
-      "required_pages_per_second_upper_estimate": 0.004366876141914569
+      "required_token_positions_per_second": 18966.901492517954,
+      "required_pages_per_second_upper_estimate": 0.004387658854228653
     },
     {
       "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": 28315.59347437504,
-      "required_pages_per_second_upper_estimate": 0.0017242710821518356
+      "required_token_positions_per_second": 28450.35223877693,
+      "required_pages_per_second_upper_estimate": 0.0017324771838793155
     },
     {
       "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": 28315.59347437504,
-      "required_pages_per_second_upper_estimate": 0.002179812768236836
+      "required_token_positions_per_second": 28450.35223877693,
+      "required_pages_per_second_upper_estimate": 0.0021901868709566286
     },
     {
       "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": 28315.59347437504,
-      "required_pages_per_second_upper_estimate": 0.35158490475650994
+      "required_token_positions_per_second": 28450.35223877693,
+      "required_pages_per_second_upper_estimate": 0.35325815760180945
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.3518057271483599,
-    "producer_p95_seconds": 0.384082982003747,
-    "gpu_update_seconds": 0.9764231155889705,
-    "producer_batches_per_second": 2.842477887172912,
-    "gpu_batches_per_second": 1.0241461760118287,
-    "producer_headroom": 2.775461114586129,
+    "producer_batch_seconds": 0.3205957231562593,
+    "producer_p95_seconds": 0.3842480040038936,
+    "gpu_update_seconds": 0.97179816151157,
+    "producer_batches_per_second": 3.119193201191262,
+    "gpu_batches_per_second": 1.029020263265948,
+    "producer_headroom": 3.031226218317057,
     "sustainable": true,
-    "buffer_seconds": 60.53823316651617,
+    "buffer_seconds": 60.25148601371734,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 457924.55451891344,
-    "upload_bytes_per_second": 5328764.108006176,
+    "generation_bytes_per_second": 428497.9780675102,
+    "upload_bytes_per_second": 5381342.4282777645,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.2599494617038,
+    "estimated_full_shards_per_hour": 91.94569117087345,
     "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": 1.4070672229972843,
-      "bytes": 1010316,
-      "pages": 29,
-      "shards": 1
-    },
     {
       "seconds": 1.181387053002254,
       "bytes": 11521167,
@@ -562,6 +556,12 @@
       "bytes": 11991791,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.5872819810028886,
+      "bytes": 6214149,
+      "pages": 24,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -328,3 +328,4 @@
 {"step": 147800, "loss": 0.6207619551569223, "lr": 0.0003, "grad_norm": 0.21111374772444552, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21629865402, "tokens_per_second": 149524.73886630964, "data_wait_s": 1.1203999747522175e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 147900, "loss": 0.617692619562149, "lr": 0.0003, "grad_norm": 0.13953698809683324, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21644514832, "tokens_per_second": 149675.14264660885, "data_wait_s": 1.4195000403560698e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 148000, "loss": 0.6084747966378927, "lr": 0.0003, "grad_norm": 0.15072616288645516, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21659252572, "tokens_per_second": 151366.57806812413, "data_wait_s": 1.7004000255838037e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 148100, "loss": 0.6290670800954103, "lr": 0.0003, "grad_norm": 0.1458833417856235, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21673815971, "tokens_per_second": 149335.9806593084, "data_wait_s": 9.645998943597078e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d78b999cab081dc42d0a1f530622de9c8417f3422fbd5cf0a6e44dad6dd03547
-size 703803
+oid sha256:f3458d0effd1707bccc4c419f34d3ed065c36430c1601fb230fb3779ac5749af
+size 705921