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": 75894.48382263062,
-      "required_pages_per_second_upper_estimate": 0.005250556490313826
+      "required_token_positions_per_second": 76879.12007042374,
+      "required_pages_per_second_upper_estimate": 0.005318675910607015
     },
     {
       "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": 9486.810477828827,
-      "required_pages_per_second_upper_estimate": 0.00397192943993664
+      "required_token_positions_per_second": 9609.890008802968,
+      "required_pages_per_second_upper_estimate": 0.004023460269362639
     },
     {
       "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": 18973.620955657654,
-      "required_pages_per_second_upper_estimate": 0.003951704877030596
+      "required_token_positions_per_second": 19219.780017605935,
+      "required_pages_per_second_upper_estimate": 0.004002973317983415
     },
     {
       "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": 28460.43143348648,
-      "required_pages_per_second_upper_estimate": 0.001706159407874932
+      "required_token_positions_per_second": 28829.670026408905,
+      "required_pages_per_second_upper_estimate": 0.001728294697725945
     },
     {
       "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": 28460.43143348648,
-      "required_pages_per_second_upper_estimate": 0.002212673359745573
+      "required_token_positions_per_second": 28829.670026408905,
+      "required_pages_per_second_upper_estimate": 0.0022413800362363623
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28460.43143348648,
-      "required_pages_per_second_upper_estimate": 0.3650598559983386
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28829.670026408905,
+      "required_pages_per_second_upper_estimate": 0.36979605221083495
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06789165754682358,
-    "producer_p95_seconds": 0.23576905200025067,
-    "gpu_update_seconds": 0.9714540014832461,
-    "producer_batches_per_second": 14.729350204924943,
-    "gpu_batches_per_second": 1.029384817472746,
-    "producer_headroom": 14.308886195822408,
+    "producer_batch_seconds": 0.11975442975008832,
+    "producer_p95_seconds": 0.24196051400213037,
+    "gpu_update_seconds": 0.9590120169489815,
+    "producer_batches_per_second": 8.350421793054904,
+    "gpu_batches_per_second": 1.0427398013024054,
+    "producer_headroom": 8.008154846132314,
     "sustainable": true,
-    "buffer_seconds": 60.23014809196126,
+    "buffer_seconds": 60.417757067785836,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 315271.23106355796,
-    "upload_bytes_per_second": 5868749.060996598,
+    "generation_bytes_per_second": 315131.5833458021,
+    "upload_bytes_per_second": 5730722.558250349,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.6498670475965,
+    "estimated_full_shards_per_hour": 67.61990189819858,
     "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": 1.1130535650008824,
-      "bytes": 11588700,
-      "pages": 34,
-      "shards": 4
-    },
     {
       "seconds": 1.0404952170010802,
       "bytes": 6993608,
@@ -562,6 +556,12 @@
       "bytes": 11210739,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 1.054491422000865,
+      "bytes": 881075,
+      "pages": 25,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -216,3 +216,4 @@
 {"step": 706600, "loss": 0.5602980852127075, "lr": 8e-05, "grad_norm": 0.20655222255051234, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103426180889, "tokens_per_second": 140863.6111176759, "data_wait_s": 1.373200211673975e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 706700, "loss": 0.5526378750801086, "lr": 8e-05, "grad_norm": 0.2174005814233793, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103440948865, "tokens_per_second": 153102.94859574814, "data_wait_s": 1.7291000403929502e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 706800, "loss": 0.5665615797042847, "lr": 8e-05, "grad_norm": 0.2775661532529385, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103455644429, "tokens_per_second": 153525.07218293217, "data_wait_s": 9.347997547592968e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 706900, "loss": 0.5981852412223816, "lr": 8e-05, "grad_norm": 0.22925572127747484, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103470381310, "tokens_per_second": 153123.57898828003, "data_wait_s": 1.2748001608997583e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a23f56efa8e49efd7b11f7a0cff52e56abb5101e3adfab4dd81c2b097329290b
-size 465021
+oid sha256:948f850eec36adb46f4a69fcd51796b02e6bd2118beb645ceb84eb65e15d9f8f
+size 467139