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": 76914.80452567649,
-      "required_pages_per_second_upper_estimate": 0.005321144644020745
+      "required_token_positions_per_second": 77250.07618813888,
+      "required_pages_per_second_upper_estimate": 0.005344339515567328
     },
     {
       "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": 9614.350565709561,
-      "required_pages_per_second_upper_estimate": 0.00402532781139241
+      "required_token_positions_per_second": 9656.25952351736,
+      "required_pages_per_second_upper_estimate": 0.0040428742168679786
     },
     {
       "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": 19228.701131419122,
-      "required_pages_per_second_upper_estimate": 0.004004831350725109
+      "required_token_positions_per_second": 19312.51904703472,
+      "required_pages_per_second_upper_estimate": 0.004022288412120758
     },
     {
       "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": 28843.05169712868,
-      "required_pages_per_second_upper_estimate": 0.0017290969084529674
+      "required_token_positions_per_second": 28968.77857055208,
+      "required_pages_per_second_upper_estimate": 0.00173663404254088
     },
     {
       "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": 28843.05169712868,
-      "required_pages_per_second_upper_estimate": 0.002242420402275076
+      "required_token_positions_per_second": 28968.77857055208,
+      "required_pages_per_second_upper_estimate": 0.0022521951136696733
     },
     {
       "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": 28843.05169712868,
-      "required_pages_per_second_upper_estimate": 0.3699676979147097
+      "required_token_positions_per_second": 28968.77857055208,
+      "required_pages_per_second_upper_estimate": 0.371580387251986
     }
   ],
   "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": 0.07252681589054077,
-    "producer_p95_seconds": 0.19940804899852083,
-    "gpu_update_seconds": 0.9585670854222527,
-    "producer_batches_per_second": 13.788003619367824,
-    "gpu_batches_per_second": 1.0432238027028604,
-    "producer_headroom": 13.216726443208886,
+    "producer_batch_seconds": 0.19729731139057094,
+    "producer_p95_seconds": 0.22782824900059495,
+    "gpu_update_seconds": 0.9544068257025271,
+    "producer_batches_per_second": 5.068492788634074,
+    "gpu_batches_per_second": 1.0477712156594357,
+    "producer_headroom": 4.837404113496396,
     "sustainable": true,
-    "buffer_seconds": 60.38972638160192,
+    "buffer_seconds": 60.12763001925921,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 174,
+    "upload_pages_per_commit": 164,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 608653.2925266916,
+    "generation_bytes_per_second": 581596.5719868013,
     "upload_bytes_per_second": 5387866.501914621,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.30141392636567,
+    "estimated_full_shards_per_hour": 62.398542736544755,
     "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-v1/1791399463488568880-054ba646/train.jsonl

@@ -108,3 +108,4 @@
 {"step": 895800, "loss": 0.5227249264717102, "lr": 3e-05, "grad_norm": 0.3118212020557827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131119945106, "tokens_per_second": 151109.3078300012, "data_wait_s": 1.5307001376640983e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 895900, "loss": 0.52464759349823, "lr": 3e-05, "grad_norm": 0.3165974995314781, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131134480203, "tokens_per_second": 152016.57642530568, "data_wait_s": 1.2111999240005389e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 896000, "loss": 0.5324521064758301, "lr": 3e-05, "grad_norm": 0.3376438801624732, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131149206160, "tokens_per_second": 153998.16058939823, "data_wait_s": 1.2835998859372921e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 896100, "loss": 0.5529004335403442, "lr": 3e-05, "grad_norm": 0.3129100120792373, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131163962648, "tokens_per_second": 154095.66123704237, "data_wait_s": 1.1449999874457717e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7a4a8b0c765a7f0f5b1d5d12dbe681edd331faf745426fc98842b51962233504
-size 234711
+oid sha256:fe122515ad2f31f2c67391017f3687a983e75e0346d40f3094eddfba5db9a8e6
+size 236829