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": 72797.82001454399,
-      "required_pages_per_second_upper_estimate": 0.0063099305344726676
+      "required_token_positions_per_second": 73870.68366151793,
+      "required_pages_per_second_upper_estimate": 0.0064029236362443155
     },
     {
       "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": 9099.727501817999,
-      "required_pages_per_second_upper_estimate": 0.00363001735352561
+      "required_token_positions_per_second": 9233.835457689742,
+      "required_pages_per_second_upper_estimate": 0.003683515022215471
     },
     {
       "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": 18199.455003635998,
-      "required_pages_per_second_upper_estimate": 0.00421012361562271
+      "required_token_positions_per_second": 18467.670915379484,
+      "required_pages_per_second_upper_estimate": 0.004272170646365741
     },
     {
       "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": 27299.182505454,
-      "required_pages_per_second_upper_estimate": 0.0016623769868407655
+      "required_token_positions_per_second": 27701.506373069227,
+      "required_pages_per_second_upper_estimate": 0.0016868763995483331
     },
     {
       "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": 27299.182505454,
-      "required_pages_per_second_upper_estimate": 0.0021015666382436533
+      "required_token_positions_per_second": 27701.506373069227,
+      "required_pages_per_second_upper_estimate": 0.0021325386432764193
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27299.182505454,
-      "required_pages_per_second_upper_estimate": 0.33896448223119807
+      "required_token_positions_per_second": 27701.506373069227,
+      "required_pages_per_second_upper_estimate": 0.34395999817561157
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 60,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.5025469781871266,
-    "producer_p95_seconds": 0.41311879999557277,
-    "gpu_update_seconds": 1.0127775802252068,
-    "producer_batches_per_second": 1.9898637210144434,
-    "gpu_batches_per_second": 0.987383626499349,
-    "producer_headroom": 2.015289364346934,
+    "producer_batch_seconds": 0.14180047160090226,
+    "producer_p95_seconds": 0.4318518479994964,
+    "gpu_update_seconds": 0.9980684670231058,
+    "producer_batches_per_second": 7.05216272350985,
+    "gpu_batches_per_second": 1.0019352710166822,
+    "producer_headroom": 7.038541238650967,
     "sustainable": true,
-    "buffer_seconds": 60.76665481351241,
+    "buffer_seconds": 60.882176488409456,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 269225.2773792399,
-    "upload_bytes_per_second": 6214279.952593561,
+    "generation_bytes_per_second": 269492.314918925,
+    "upload_bytes_per_second": 6265484.452232233,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.76947728188417,
+    "estimated_full_shards_per_hour": 57.82677732158482,
     "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": 0.828995023999596,
-      "bytes": 4743589,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 0.9685989429999609,
       "bytes": 7522789,
@@ -562,6 +556,12 @@
       "bytes": 919630,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.1119171959944651,
+      "bytes": 9959227,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -406,3 +406,4 @@
 {"step": 80600, "loss": 0.6215187516063452, "lr": 0.0003, "grad_norm": 0.13800425708100492, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11795271071, "tokens_per_second": 149596.1864688307, "data_wait_s": 8.9899986051023e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 80700, "loss": 0.6559007130563259, "lr": 0.0003, "grad_norm": 0.14497893489856906, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11810066328, "tokens_per_second": 149685.25251918632, "data_wait_s": 1.3229000614956021e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 80800, "loss": 0.6198956929147243, "lr": 0.0003, "grad_norm": 0.14106941438407244, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11824774435, "tokens_per_second": 149304.25993980162, "data_wait_s": 9.3669950729236e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 80900, "loss": 0.6130074709653854, "lr": 0.0003, "grad_norm": 0.1471136870582393, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11839396826, "tokens_per_second": 147751.93799730256, "data_wait_s": 1.2246004189364612e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f354e828348cef99144b296db9f27dfe61d01d5de54d2e3dbc08c6dfafd933c1
-size 870573
+oid sha256:00f503a41e8c1144a950e4478e1dade0c069c947eb6351bd50005f690130f88b
+size 872691