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": 75517.98712793214,
-      "required_pages_per_second_upper_estimate": 0.005224509575381896
+      "required_token_positions_per_second": 76254.62031560014,
+      "required_pages_per_second_upper_estimate": 0.005275471568529249
     },
     {
       "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": 9439.748390991517,
-      "required_pages_per_second_upper_estimate": 0.003952225526945986
+      "required_token_positions_per_second": 9531.827539450018,
+      "required_pages_per_second_upper_estimate": 0.003990777143574288
     },
     {
       "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": 18879.496781983034,
-      "required_pages_per_second_upper_estimate": 0.0039321012938755544
+      "required_token_positions_per_second": 19063.655078900036,
+      "required_pages_per_second_upper_estimate": 0.003970456610542485
     },
     {
       "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": 28319.24517297455,
-      "required_pages_per_second_upper_estimate": 0.0016976955071361793
+      "required_token_positions_per_second": 28595.48261835005,
+      "required_pages_per_second_upper_estimate": 0.001714255520196284
     },
     {
       "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": 28319.24517297455,
-      "required_pages_per_second_upper_estimate": 0.0022016967490035
+      "required_token_positions_per_second": 28595.48261835005,
+      "required_pages_per_second_upper_estimate": 0.002223172995341332
     },
     {
       "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": 28319.24517297455,
-      "required_pages_per_second_upper_estimate": 0.3632488702424873
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28595.48261835005,
+      "required_pages_per_second_upper_estimate": 0.3667921475911039
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.07550128316406735,
-    "producer_p95_seconds": 0.1326797340000212,
-    "gpu_update_seconds": 0.9762972081750565,
-    "producer_batches_per_second": 13.244808009778582,
-    "gpu_batches_per_second": 1.0242782542308504,
-    "producer_headroom": 12.930869082761456,
+    "producer_batch_seconds": 0.05795549093751262,
+    "producer_p95_seconds": 0.130428725999991,
+    "gpu_update_seconds": 0.9668660035923982,
+    "producer_batches_per_second": 17.254620465180704,
+    "gpu_batches_per_second": 1.0342694812771287,
+    "producer_headroom": 16.68290593267287,
     "sustainable": true,
-    "buffer_seconds": 60.530426906853506,
+    "buffer_seconds": 60.91255822632108,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 62,
+    "upload_pages_per_commit": 63,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 374018.94871161104,
+    "generation_bytes_per_second": 409663.89748946467,
     "upload_bytes_per_second": 4309465.406494768,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.25575967799423,
+    "estimated_full_shards_per_hour": 87.9043359137817,
     "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-v1/1790989041640695387-36510d53/train.jsonl

@@ -7,3 +7,4 @@
 {"step": 557800, "loss": 0.6057742834091187, "lr": 0.0003, "grad_norm": 0.15931434321151466, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81644154641, "tokens_per_second": 150808.6348368384, "data_wait_s": 1.4784999848416192e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 557900, "loss": 0.5840400457382202, "lr": 0.0003, "grad_norm": 0.19269194496021208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81658761408, "tokens_per_second": 152351.4797034027, "data_wait_s": 9.032999969349476e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 558000, "loss": 0.5602437853813171, "lr": 0.0003, "grad_norm": 0.1694346849309562, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81673385598, "tokens_per_second": 151518.40913651977, "data_wait_s": 1.511400000708818e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 558100, "loss": 0.597959041595459, "lr": 0.0003, "grad_norm": 0.1546086724713371, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81687934063, "tokens_per_second": 151342.1559679288, "data_wait_s": 1.2883999943369417e-05, "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 sha256:1107c916497863d563e267ff12096299b7e5a47408e5a733aa121c1182597906
-size 19227
+oid sha256:c5cdb3926ca2bd43020c8bd7ba2586425718ebbf39cdc4a020d8356673c0290e
+size 21345