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": 75057.26276596382,
-      "required_pages_per_second_upper_estimate": 0.006505773305660439
+      "required_token_positions_per_second": 74418.44499214872,
+      "required_pages_per_second_upper_estimate": 0.006450402200094992
     },
     {
       "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": 9382.157845745478,
-      "required_pages_per_second_upper_estimate": 0.0037426830404282264
+      "required_token_positions_per_second": 9302.30562401859,
+      "required_pages_per_second_upper_estimate": 0.003710828795284263
     },
     {
       "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": 18764.315691490956,
-      "required_pages_per_second_upper_estimate": 0.004340794194549382
+      "required_token_positions_per_second": 18604.61124803718,
+      "required_pages_per_second_upper_estimate": 0.004303849382258558
     },
     {
       "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": 28146.47353723643,
-      "required_pages_per_second_upper_estimate": 0.0017139725652838297
+      "required_token_positions_per_second": 27906.91687205577,
+      "required_pages_per_second_upper_estimate": 0.001699384821231015
     },
     {
       "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": 28146.47353723643,
-      "required_pages_per_second_upper_estimate": 0.0021667934473219566
+      "required_token_positions_per_second": 27906.91687205577,
+      "required_pages_per_second_upper_estimate": 0.0021483517121010575
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28146.47353723643,
-      "required_pages_per_second_upper_estimate": 0.34948500114526776
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27906.91687205577,
+      "required_pages_per_second_upper_estimate": 0.3465105091083076
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.12556662605476276,
-    "producer_p95_seconds": 0.42661202199815307,
-    "gpu_update_seconds": 0.9822900180878086,
-    "producer_batches_per_second": 7.963899576021696,
-    "gpu_batches_per_second": 1.018029280137313,
-    "producer_headroom": 7.8228590585798425,
+    "producer_batch_seconds": 0.12618614066366263,
+    "producer_p95_seconds": 0.43406799800141016,
+    "gpu_update_seconds": 0.9907221255130826,
+    "producer_batches_per_second": 7.924800574299254,
+    "gpu_batches_per_second": 1.0093647595506283,
+    "producer_headroom": 7.851275269237054,
     "sustainable": true,
-    "buffer_seconds": 60.90198112144413,
+    "buffer_seconds": 60.434049656298036,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 483536.875749722,
+    "generation_bytes_per_second": 486334.2840626001,
     "upload_bytes_per_second": 5773531.969764306,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.75575737351174,
+    "estimated_full_shards_per_hour": 104.35601607712272,
     "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": [

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

@@ -404,3 +404,4 @@
 {"step": 155400, "loss": 0.619165938347578, "lr": 0.0003, "grad_norm": 0.1476237453690166, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22742793235, "tokens_per_second": 148852.73737805066, "data_wait_s": 1.5402001736219972e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 155500, "loss": 0.5956929791718721, "lr": 0.0003, "grad_norm": 0.1505320523732815, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22757469182, "tokens_per_second": 149098.32961767094, "data_wait_s": 1.2772994523402303e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 155600, "loss": 0.6896321140229702, "lr": 0.0003, "grad_norm": 0.16904787321003917, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22772124076, "tokens_per_second": 149032.15146045198, "data_wait_s": 1.2080003216397017e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 155700, "loss": 0.5687103159725666, "lr": 0.0003, "grad_norm": 0.14221636869850315, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22786637208, "tokens_per_second": 147880.678066207, "data_wait_s": 1.4651996025349945e-05, "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 sha256:2018aee8ecfd820d1f136f3e212952e9843e770dccfdd81ee2654877d64db5c3
-size 866337
+oid sha256:a5c46f725e92021e6e8dafdf65c22ade4da85a394692562fc1b3122e2ee97815
+size 868455