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": 76946.25313861176,
-      "required_pages_per_second_upper_estimate": 0.005323320332034442
+      "required_token_positions_per_second": 77211.01571349954,
+      "required_pages_per_second_upper_estimate": 0.005341637221298991
     },
     {
       "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": 9618.28164232647,
-      "required_pages_per_second_upper_estimate": 0.004026973671081695
+      "required_token_positions_per_second": 9651.376964187442,
+      "required_pages_per_second_upper_estimate": 0.004040829991235975
     },
     {
       "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": 19236.56328465294,
-      "required_pages_per_second_upper_estimate": 0.00400646882990479
+      "required_token_positions_per_second": 19302.753928374885,
+      "required_pages_per_second_upper_estimate": 0.004020254595427411
     },
     {
       "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": 28854.84492697941,
-      "required_pages_per_second_upper_estimate": 0.0017298038945752993
+      "required_token_positions_per_second": 28954.13089256233,
+      "required_pages_per_second_upper_estimate": 0.0017357559366111038
     },
     {
       "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": 28854.84492697941,
-      "required_pages_per_second_upper_estimate": 0.002243337274023034
+      "required_token_positions_per_second": 28954.13089256233,
+      "required_pages_per_second_upper_estimate": 0.0022510563211342966
     },
     {
       "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": 28854.84492697941,
-      "required_pages_per_second_upper_estimate": 0.37011896880465117
+      "required_token_positions_per_second": 28954.13089256233,
+      "required_pages_per_second_upper_estimate": 0.3713925025661848
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0856779190702639,
-    "producer_p95_seconds": 0.19173568499900284,
-    "gpu_update_seconds": 0.9581753105922601,
-    "producer_batches_per_second": 11.671618672016375,
-    "gpu_batches_per_second": 1.0436503518149383,
-    "producer_headroom": 11.183456846173714,
+    "producer_batch_seconds": 0.07493022167987817,
+    "producer_p95_seconds": 0.192103995999787,
+    "gpu_update_seconds": 0.9548896529683837,
+    "producer_batches_per_second": 13.345749920135908,
+    "gpu_batches_per_second": 1.0472414240654777,
+    "producer_headroom": 12.743718509841411,
     "sustainable": true,
-    "buffer_seconds": 60.36504456731239,
+    "buffer_seconds": 60.15804813700817,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 441899.53516029747,
-    "upload_bytes_per_second": 3474730.911396226,
+    "generation_bytes_per_second": 465170.82398095325,
+    "upload_bytes_per_second": 3508263.1892297296,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.82135335070318,
+    "estimated_full_shards_per_hour": 99.81483020373771,
     "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": [
@@ -394,6 +394,12 @@
       "bytes": 8208196,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.0678595420013153,
+      "bytes": 6041574,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -85,3 +85,4 @@
 {"step": 693500, "loss": 0.5474476218223572, "lr": 8e-05, "grad_norm": 0.20007133977990868, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101509582544, "tokens_per_second": 151203.60353158307, "data_wait_s": 1.8898999769589864e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 693600, "loss": 0.5194411277770996, "lr": 8e-05, "grad_norm": 0.19325535197729066, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101524352815, "tokens_per_second": 153224.88824170307, "data_wait_s": 1.509200046712067e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 693700, "loss": 0.4594331383705139, "lr": 8e-05, "grad_norm": 0.19222642154230435, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101538986975, "tokens_per_second": 152510.79430010437, "data_wait_s": 1.909300044644624e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 693800, "loss": 0.5116367340087891, "lr": 8e-05, "grad_norm": 0.2069830763113695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101553667688, "tokens_per_second": 153496.04986118193, "data_wait_s": 1.867600076366216e-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:f1095dcd08c458be3084af56c6d44bac6b0a5662000aa713e8c02078430bb77f
-size 185214
+oid sha256:60c629949113d2349b8d0cbb35e636b02fa03a4cfc40188bae5d0fda143b6688
+size 187332