Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77043.62593056588,
-      "required_pages_per_second_upper_estimate": 0.005330056807717832
+      "required_token_positions_per_second": 76357.4890183843,
+      "required_pages_per_second_upper_estimate": 0.005282588264076132
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9630.453241320734,
-      "required_pages_per_second_upper_estimate": 0.004032069665408704
+      "required_token_positions_per_second": 9544.686127298037,
+      "required_pages_per_second_upper_estimate": 0.003996160765788405
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19260.90648264147,
-      "required_pages_per_second_upper_estimate": 0.004011538876072499
+      "required_token_positions_per_second": 19089.372254596074,
+      "required_pages_per_second_upper_estimate": 0.003975812820032439
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28891.359723962207,
-      "required_pages_per_second_upper_estimate": 0.0017319929008995513
+      "required_token_positions_per_second": 28634.058381894112,
+      "required_pages_per_second_upper_estimate": 0.00171656807831897
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28891.359723962207,
-      "required_pages_per_second_upper_estimate": 0.0022461761388768283
+      "required_token_positions_per_second": 28634.058381894112,
+      "required_pages_per_second_upper_estimate": 0.002226172091280031
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28891.359723962207,
-      "required_pages_per_second_upper_estimate": 0.3705873414138121
+      "required_token_positions_per_second": 28634.058381894112,
+      "required_pages_per_second_upper_estimate": 0.36728695606641926
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11282432164819056,
-    "producer_p95_seconds": 0.33012011000391794,
-    "gpu_update_seconds": 0.9569643057356357,
-    "producer_batches_per_second": 8.863337136811738,
-    "gpu_batches_per_second": 1.0449710548308089,
-    "producer_headroom": 8.481897269629924,
+    "producer_batch_seconds": 0.45491830740644446,
+    "producer_p95_seconds": 0.3204213319986593,
+    "gpu_update_seconds": 0.9655634430599046,
+    "producer_batches_per_second": 2.1981968712166053,
+    "gpu_batches_per_second": 1.0356647273543877,
+    "producer_headroom": 2.1224985394954152,
     "sustainable": true,
-    "buffer_seconds": 60.28875126134505,
+    "buffer_seconds": 60.83049691277399,
     "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": 335173.44269849313,
-    "upload_bytes_per_second": 4993552.828645275,
+    "generation_bytes_per_second": 305820.9735257099,
+    "upload_bytes_per_second": 4967357.71113432,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.92041836467834,
+    "estimated_full_shards_per_hour": 65.62206177071069,
     "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": 1.1605264330028149,
-      "bytes": 9613122,
-      "pages": 38,
-      "shards": 4
-    },
     {
       "seconds": 1.0401736989988422,
       "bytes": 6081990,
@@ -562,6 +556,12 @@
       "bytes": 11122347,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.0296640300002764,
+      "bytes": 6697132,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -349,3 +349,4 @@
 {"step": 592000, "loss": 0.6148668527603149, "lr": 0.0003, "grad_norm": 0.19262962241393378, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86650986318, "tokens_per_second": 152141.1724714603, "data_wait_s": 1.084299583453685e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 592100, "loss": 0.5621565580368042, "lr": 0.0003, "grad_norm": 0.16573701637922625, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86665577589, "tokens_per_second": 151326.54263661295, "data_wait_s": 1.1258998711127788e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 592200, "loss": 0.5876827239990234, "lr": 0.0003, "grad_norm": 0.1504776926430603, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86680156809, "tokens_per_second": 152306.231422966, "data_wait_s": 9.504001354798675e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 592300, "loss": 0.60694819688797, "lr": 0.0003, "grad_norm": 0.1776004738602595, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86694996615, "tokens_per_second": 133624.01058406642, "data_wait_s": 9.276001947000623e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f93c3b716ddd059effe4762f99ff8945cbfb94ba60ad0791f4c7e6a1c99bddff
-size 749064
+oid sha256:e5b06e6cb1f5b8177d9782c1c518dede3687d3cdc0e4d841dfd0e9ee58f59b0d
+size 751182