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": 75342.59954379244,
-      "required_pages_per_second_upper_estimate": 0.006530505574383172
+      "required_token_positions_per_second": 75583.06335444357,
+      "required_pages_per_second_upper_estimate": 0.006551348368040456
     },
     {
       "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": 9417.824942974055,
-      "required_pages_per_second_upper_estimate": 0.0037569111787833314
+      "required_token_positions_per_second": 9447.882919305446,
+      "required_pages_per_second_upper_estimate": 0.003768901754948718
     },
     {
       "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": 18835.64988594811,
-      "required_pages_per_second_upper_estimate": 0.004357296104998101
+      "required_token_positions_per_second": 18895.765838610892,
+      "required_pages_per_second_upper_estimate": 0.004371202872642004
     },
     {
       "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": 28253.474828922164,
-      "required_pages_per_second_upper_estimate": 0.0017204883825551003
+      "required_token_positions_per_second": 28343.648757916337,
+      "required_pages_per_second_upper_estimate": 0.0017259795017248035
     },
     {
       "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": 28253.474828922164,
-      "required_pages_per_second_upper_estimate": 0.0021750307029544573
+      "required_token_positions_per_second": 28343.648757916337,
+      "required_pages_per_second_upper_estimate": 0.0021819725416258404
     },
     {
       "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": 28253.474828922164,
-      "required_pages_per_second_upper_estimate": 0.35081359907771786
+      "required_token_positions_per_second": 28343.648757916337,
+      "required_pages_per_second_upper_estimate": 0.3519332574830989
     }
   ],
   "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": 0.3277401188977933,
-    "producer_p95_seconds": 0.40702690499892924,
-    "gpu_update_seconds": 0.9785698986553555,
-    "producer_batches_per_second": 3.0511980143384667,
-    "gpu_batches_per_second": 1.021899407874789,
-    "producer_headroom": 2.9858105316686157,
+    "producer_batch_seconds": 0.11979209478852226,
+    "producer_p95_seconds": 0.38984344500204315,
+    "gpu_update_seconds": 0.975456626496543,
+    "producer_batches_per_second": 8.34779625287773,
+    "gpu_batches_per_second": 1.0251609070426915,
+    "producer_headroom": 8.142913171512593,
     "sustainable": true,
-    "buffer_seconds": 60.67133371663204,
+    "buffer_seconds": 60.47831084278566,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 335703.3398410713,
-    "upload_bytes_per_second": 4037513.2741344315,
+    "generation_bytes_per_second": 379734.8500496105,
+    "upload_bytes_per_second": 4092962.040476732,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.03412195610146,
+    "estimated_full_shards_per_hour": 81.48225904575573,
     "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": 1.141221003003011,
-      "bytes": 7503350,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.1287752040007035,
       "bytes": 10564318,
@@ -562,6 +556,12 @@
       "bytes": 6127988,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.1608434069930809,
+      "bytes": 13413721,
+      "pages": 38,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -347,3 +347,4 @@
 {"step": 349700, "loss": 0.5897213071584702, "lr": 0.0003, "grad_norm": 0.149293986108801, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51187014457, "tokens_per_second": 149374.86428809058, "data_wait_s": 1.1515003279782832e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 349800, "loss": 0.5988677442073822, "lr": 0.0003, "grad_norm": 0.17552263591675815, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51201641908, "tokens_per_second": 148145.49559771136, "data_wait_s": 1.6423997294623405e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 349900, "loss": 0.584590258076787, "lr": 0.0003, "grad_norm": 0.17714293403909717, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51216225970, "tokens_per_second": 147863.93379601528, "data_wait_s": 9.56900476012379e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 350000, "loss": 0.6504786228761077, "lr": 0.0003, "grad_norm": 0.1511979477375556, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51230919415, "tokens_per_second": 148850.98533615225, "data_wait_s": 1.4363999071065336e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:91e23be9af116b11bd54d2edcd5948228d0de235fd4245efa1cd7f07569e8f97
-size 744045
+oid sha256:2805347b6cdd667f4bbe282700fec9a8a123aec693239cc67e63a12bfc26a12d
+size 746163