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": 75496.40011300516,
-      "required_pages_per_second_upper_estimate": 0.006543836617918552
+      "required_token_positions_per_second": 74766.12188342438,
+      "required_pages_per_second_upper_estimate": 0.00648053795185174
     },
     {
       "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": 9437.050014125645,
-      "required_pages_per_second_upper_estimate": 0.0037645803471061294
+      "required_token_positions_per_second": 9345.765235428047,
+      "required_pages_per_second_upper_estimate": 0.0037281654840545905
     },
     {
       "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": 18874.10002825129,
-      "required_pages_per_second_upper_estimate": 0.004366190868720549
+      "required_token_positions_per_second": 18691.530470856094,
+      "required_pages_per_second_upper_estimate": 0.004323956614731639
     },
     {
       "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": 28311.150042376936,
-      "required_pages_per_second_upper_estimate": 0.001724000500456037
+      "required_token_positions_per_second": 28037.29570628414,
+      "required_pages_per_second_upper_estimate": 0.0017073242081906451
     },
     {
       "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": 28311.150042376936,
-      "required_pages_per_second_upper_estimate": 0.002179470700541415
+      "required_token_positions_per_second": 28037.29570628414,
+      "required_pages_per_second_upper_estimate": 0.00215838863567167
     },
     {
       "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": 28311.150042376936,
-      "required_pages_per_second_upper_estimate": 0.3515297322023037
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28037.29570628414,
+      "required_pages_per_second_upper_estimate": 0.3481293778795354
     }
   ],
   "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": 0.086718972453113,
-    "producer_p95_seconds": 0.2092034170000261,
-    "gpu_update_seconds": 0.9765763650934591,
-    "producer_batches_per_second": 11.531501950633439,
-    "gpu_batches_per_second": 1.023985461602175,
-    "producer_headroom": 11.261392259017738,
+    "producer_batch_seconds": 0.4761266152577832,
+    "producer_p95_seconds": 0.30677174199990986,
+    "gpu_update_seconds": 0.9861150764908869,
+    "producer_batches_per_second": 2.1002816644866256,
+    "gpu_batches_per_second": 1.0140804291914114,
+    "producer_headroom": 2.0711194142276357,
     "sustainable": true,
-    "buffer_seconds": 60.54773463579447,
+    "buffer_seconds": 60.1530196659441,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 256200.76188808383,
+    "generation_bytes_per_second": 237496.29901323313,
     "upload_bytes_per_second": 6248432.595319933,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.97471945268522,
+    "estimated_full_shards_per_hour": 50.961177137353374,
     "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-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -143,3 +143,4 @@
 {"step": 14300, "loss": 0.6647667810320854, "lr": 0.0003, "grad_norm": 0.14701070813532818, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2092167327, "tokens_per_second": 148300.1718091062, "data_wait_s": 1.3512999430531636e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 14400, "loss": 0.6537965591996908, "lr": 0.0003, "grad_norm": 0.13906760496417342, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2106797171, "tokens_per_second": 149334.5009564062, "data_wait_s": 1.3601998944068328e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 14500, "loss": 0.6891657430678606, "lr": 0.0003, "grad_norm": 0.14184640324441833, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2121506262, "tokens_per_second": 149796.43585675984, "data_wait_s": 1.0572999599389732e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 14600, "loss": 0.6604617889970541, "lr": 0.0003, "grad_norm": 0.1363267703414383, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2136234563, "tokens_per_second": 148322.30530545712, "data_wait_s": 8.441998943453655e-06, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:15b2678feb6b1f93258456dd46adb3d339dc0af024124aeb7bceb0313e44c9f8
-size 303707
+oid sha256:3d3a1ff174394d08049a091ad96e62ec0e2007012e787e75fe4f772ee0f00dee
+size 305790