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": 75023.99834861016,
-      "required_pages_per_second_upper_estimate": 0.006502890030272118
+      "required_token_positions_per_second": 75608.98958113827,
+      "required_pages_per_second_upper_estimate": 0.006553595587660932
     },
     {
       "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": 9377.99979357627,
-      "required_pages_per_second_upper_estimate": 0.003741024331249509
+      "required_token_positions_per_second": 9451.123697642284,
+      "required_pages_per_second_upper_estimate": 0.0037701945498812366
     },
     {
       "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": 18755.99958715254,
-      "required_pages_per_second_upper_estimate": 0.004338870410169127
+      "required_token_positions_per_second": 18902.24739528457,
+      "required_pages_per_second_upper_estimate": 0.004372702266706958
     },
     {
       "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": 28133.999380728805,
-      "required_pages_per_second_upper_estimate": 0.0017132129545993566
+      "required_token_positions_per_second": 28353.371092926853,
+      "required_pages_per_second_upper_estimate": 0.0017265715409177944
     },
     {
       "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": 28133.999380728805,
-      "required_pages_per_second_upper_estimate": 0.0021658331522233245
+      "required_token_positions_per_second": 28353.371092926853,
+      "required_pages_per_second_upper_estimate": 0.002182720993888096
     },
     {
       "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": 28133.999380728805,
-      "required_pages_per_second_upper_estimate": 0.3493301138697593
+      "required_token_positions_per_second": 28353.371092926853,
+      "required_pages_per_second_upper_estimate": 0.3520539763453674
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10496264092969909,
-    "producer_p95_seconds": 0.3569807860003493,
-    "gpu_update_seconds": 0.9827255494623454,
-    "producer_batches_per_second": 9.527199307701974,
-    "gpu_batches_per_second": 1.0175781026015918,
-    "producer_headroom": 9.362622174498698,
+    "producer_batch_seconds": 0.44117724729693464,
+    "producer_p95_seconds": 0.35217179200117243,
+    "gpu_update_seconds": 0.9751221436556863,
+    "producer_batches_per_second": 2.266662676116997,
+    "gpu_batches_per_second": 1.0255125539976437,
+    "producer_headroom": 2.2102729676795407,
     "sustainable": true,
-    "buffer_seconds": 60.928984066665414,
+    "buffer_seconds": 60.45757290665255,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 324116.32493049925,
+    "generation_bytes_per_second": 284533.2303471937,
     "upload_bytes_per_second": 3837682.50732972,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.54781828819497,
+    "estimated_full_shards_per_hour": 61.05420763789996,
     "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

@@ -265,3 +265,4 @@
 {"step": 141500, "loss": 0.5589583292603493, "lr": 0.0003, "grad_norm": 0.12546013281206278, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20708449624, "tokens_per_second": 149931.85366369397, "data_wait_s": 1.1326999810989946e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 141600, "loss": 0.6302874404937029, "lr": 0.0003, "grad_norm": 0.14658013483266016, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20723265050, "tokens_per_second": 150394.1371699987, "data_wait_s": 1.3592998584499583e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 141700, "loss": 0.5687056966125965, "lr": 0.0003, "grad_norm": 0.14378563050977897, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20737973026, "tokens_per_second": 150134.11335283951, "data_wait_s": 1.2888998753624037e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 141800, "loss": 0.5475311782211065, "lr": 0.0003, "grad_norm": 0.13591897510819098, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20752416941, "tokens_per_second": 146928.80135276602, "data_wait_s": 9.723000403027982e-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 sha256:7e141659164c818836b9e453d0971b2766120d4e39242ff6a4425b44a7e72158
-size 569586
+oid sha256:deaa5ec6554df0248de13b055c51ba400524fcbc9f192a4a1108e7b80bc41236
+size 571704