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": 76249.90983229561,
-      "required_pages_per_second_upper_estimate": 0.005275145686364414
+      "required_token_positions_per_second": 75698.87529755497,
+      "required_pages_per_second_upper_estimate": 0.005237023838674783
     },
     {
       "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": 9531.238729036952,
-      "required_pages_per_second_upper_estimate": 0.003990530620950097
+      "required_token_positions_per_second": 9462.359412194372,
+      "required_pages_per_second_upper_estimate": 0.003961692289351806
     },
     {
       "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": 19062.477458073903,
-      "required_pages_per_second_upper_estimate": 0.003970211343180352
+      "required_token_positions_per_second": 18924.718824388743,
+      "required_pages_per_second_upper_estimate": 0.003941519852723207
     },
     {
       "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": 28593.71618711086,
-      "required_pages_per_second_upper_estimate": 0.0017141496253406786
+      "required_token_positions_per_second": 28387.078236583115,
+      "required_pages_per_second_upper_estimate": 0.0017017619957244217
     },
     {
       "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": 28593.71618711086,
-      "required_pages_per_second_upper_estimate": 0.0022230356630821946
+      "required_token_positions_per_second": 28387.078236583115,
+      "required_pages_per_second_upper_estimate": 0.002206970471332951
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28593.71618711086,
-      "required_pages_per_second_upper_estimate": 0.36676948970781365
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28387.078236583115,
+      "required_pages_per_second_upper_estimate": 0.3641189599489888
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09386789268759799,
-    "producer_p95_seconds": 0.2205734189992654,
-    "gpu_update_seconds": 0.966925733579983,
-    "producer_batches_per_second": 10.653269945327343,
-    "gpu_batches_per_second": 1.0342055912583499,
-    "producer_headroom": 10.300920856911228,
+    "producer_batch_seconds": 0.43281881867193306,
+    "producer_p95_seconds": 0.25115406599798007,
+    "gpu_update_seconds": 0.9739642723909977,
+    "producer_batches_per_second": 2.3104355838048196,
+    "gpu_batches_per_second": 1.0267317070523407,
+    "producer_headroom": 2.250281712286731,
     "sustainable": true,
-    "buffer_seconds": 60.91632121553893,
+    "buffer_seconds": 60.38578488824186,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 381841.5921043337,
+    "generation_bytes_per_second": 350431.9243617246,
     "upload_bytes_per_second": 3825939.046495481,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.93431684825428,
+    "estimated_full_shards_per_hour": 75.19453333033374,
     "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-v1/1791300814123050567-72be911d/train.jsonl

@@ -164,3 +164,4 @@
 {"step": 816400, "loss": 0.4969991147518158, "lr": 3e-05, "grad_norm": 0.27524534196188283, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119497719635, "tokens_per_second": 150695.83704297701, "data_wait_s": 1.1690000974340364e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 816500, "loss": 0.5110582709312439, "lr": 3e-05, "grad_norm": 0.2823801772473911, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119512226158, "tokens_per_second": 149414.69378761944, "data_wait_s": 1.541700112284161e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 816600, "loss": 0.5242553353309631, "lr": 3e-05, "grad_norm": 0.281920959251156, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119526832504, "tokens_per_second": 151121.00753663926, "data_wait_s": 1.315700137638487e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 816700, "loss": 0.5188537836074829, "lr": 3e-05, "grad_norm": 0.2697688917024653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119541401067, "tokens_per_second": 150342.95871078558, "data_wait_s": 6.4069972722791135e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2181e142a97c5daef7dae87f29105bb63b98e30050a6fb6594cb7c473f0d75bc
-size 354102
+oid sha256:582f79723c3294def6a7f46fc6eeca4ee93b4a96de3455a8248c8c9928a46475
+size 356220