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": 76172.48875489332,
-      "required_pages_per_second_upper_estimate": 0.005269789516588075
+      "required_token_positions_per_second": 77102.9347819107,
+      "required_pages_per_second_upper_estimate": 0.005334159931669363
     },
     {
       "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": 9521.561094361665,
-      "required_pages_per_second_upper_estimate": 0.003986478797403547
+      "required_token_positions_per_second": 9637.866847738838,
+      "required_pages_per_second_upper_estimate": 0.004035173587602257
     },
     {
       "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": 19043.12218872333,
-      "required_pages_per_second_upper_estimate": 0.003966180151007405
+      "required_token_positions_per_second": 19275.733695477677,
+      "required_pages_per_second_upper_estimate": 0.004014626993486383
     },
     {
       "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": 28564.683283084996,
-      "required_pages_per_second_upper_estimate": 0.0017124091470750058
+      "required_token_positions_per_second": 28913.600543216515,
+      "required_pages_per_second_upper_estimate": 0.0017333262040540823
     },
     {
       "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": 28564.683283084996,
-      "required_pages_per_second_upper_estimate": 0.0022207784824964326
+      "required_token_positions_per_second": 28913.600543216515,
+      "required_pages_per_second_upper_estimate": 0.0022479052647468257
     },
     {
       "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": 28564.683283084996,
-      "required_pages_per_second_upper_estimate": 0.3663970867880735
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28913.600543216515,
+      "required_pages_per_second_upper_estimate": 0.3708726227628752
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10938445195323254,
-    "producer_p95_seconds": 0.22189596700081893,
-    "gpu_update_seconds": 0.9679085087693647,
-    "producer_batches_per_second": 9.142067104999084,
-    "gpu_batches_per_second": 1.0331555006902848,
-    "producer_headroom": 8.848684538669126,
+    "producer_batch_seconds": 0.05493818685165763,
+    "producer_p95_seconds": 0.18927947399970435,
+    "gpu_update_seconds": 0.9562281929805023,
+    "producer_batches_per_second": 18.20227527166429,
+    "gpu_batches_per_second": 1.0457754826105512,
+    "producer_headroom": 17.405528791157227,
     "sustainable": true,
-    "buffer_seconds": 60.01032754370061,
+    "buffer_seconds": 60.242376157771645,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 292467.8204815524,
-    "upload_bytes_per_second": 5553560.1583004,
+    "generation_bytes_per_second": 292614.0560326726,
+    "upload_bytes_per_second": 5487441.673117563,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.75678597292832,
+    "estimated_full_shards_per_hour": 62.78816471800931,
     "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": [
@@ -466,6 +466,12 @@
       "bytes": 13516314,
       "pages": 40,
       "shards": 4
+    },
+    {
+      "seconds": 0.855230520999612,
+      "bytes": 838138,
+      "pages": 29,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -94,3 +94,4 @@
 {"step": 769400, "loss": 0.48373109102249146, "lr": 3e-05, "grad_norm": 0.23279685805445827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112617267281, "tokens_per_second": 148437.4568293413, "data_wait_s": 1.5268000424839556e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 769500, "loss": 0.523493766784668, "lr": 3e-05, "grad_norm": 0.236160730364222, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112631844636, "tokens_per_second": 151244.69218965745, "data_wait_s": 1.0974999895552173e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 769600, "loss": 0.521182119846344, "lr": 3e-05, "grad_norm": 0.2844613332504851, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112646639597, "tokens_per_second": 154169.4638959554, "data_wait_s": 9.65899926086422e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 769700, "loss": 0.5391861200332642, "lr": 3e-05, "grad_norm": 0.24925319598414888, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112661368981, "tokens_per_second": 129550.51739029269, "data_wait_s": 1.6445999790448695e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5cb24afe0ae8205f07d4f4ae892ec88be4f9eccd19f05149d53be90028bae374
-size 204276
+oid sha256:b36f3049758f5eab22c19e0684b043598eb23b2b71cebde4be64370eb5dce522
+size 206394