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": 76249.61529320957,
-      "required_pages_per_second_upper_estimate": 0.005275125309467017
+      "required_token_positions_per_second": 77652.48672920754,
+      "required_pages_per_second_upper_estimate": 0.005372179210519554
     },
     {
       "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": 9531.201911651196,
-      "required_pages_per_second_upper_estimate": 0.003990515206279515
+      "required_token_positions_per_second": 9706.560841150942,
+      "required_pages_per_second_upper_estimate": 0.004063934328150201
     },
     {
       "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": 19062.403823302393,
-      "required_pages_per_second_upper_estimate": 0.003970196006999325
+      "required_token_positions_per_second": 19413.121682301884,
+      "required_pages_per_second_upper_estimate": 0.0040432412879770055
     },
     {
       "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": 28593.605734953588,
-      "required_pages_per_second_upper_estimate": 0.001714143003902501
+      "required_token_positions_per_second": 29119.682523452826,
+      "required_pages_per_second_upper_estimate": 0.0017456805041002853
     },
     {
       "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": 28593.605734953588,
-      "required_pages_per_second_upper_estimate": 0.0022230270759127945
+      "required_token_positions_per_second": 29119.682523452826,
+      "required_pages_per_second_upper_estimate": 0.0022639272322513655
     },
     {
       "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": 28593.605734953588,
-      "required_pages_per_second_upper_estimate": 0.36676807294613445
+      "required_token_positions_per_second": 29119.682523452826,
+      "required_pages_per_second_upper_estimate": 0.37351602113175597
     }
   ],
   "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": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07762395034384895,
+    "producer_batch_seconds": 0.09389047067969614,
     "producer_p95_seconds": 0.2729683910001768,
-    "gpu_update_seconds": 0.966929468646983,
-    "producer_batches_per_second": 12.882621865678365,
-    "gpu_batches_per_second": 1.034201596316319,
-    "producer_headroom": 12.456586715360386,
+    "gpu_update_seconds": 0.9494609008093566,
+    "producer_batches_per_second": 10.650708136414215,
+    "gpu_batches_per_second": 1.0532292579373852,
+    "producer_headroom": 10.112430941457385,
     "sustainable": true,
-    "buffer_seconds": 60.91655652475993,
+    "buffer_seconds": 60.765497651798825,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 265700.6272989159,
+    "generation_bytes_per_second": 278326.7992725558,
     "upload_bytes_per_second": 4679545.0310742315,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.01316942430122,
+    "estimated_full_shards_per_hour": 59.7224520076037,
     "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

@@ -295,3 +295,4 @@
 {"step": 914500, "loss": 0.5214608907699585, "lr": 3e-05, "grad_norm": 0.3191385382969053, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133856084120, "tokens_per_second": 151598.96451312004, "data_wait_s": 1.6672001947881654e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 914600, "loss": 0.4910697340965271, "lr": 3e-05, "grad_norm": 0.3203844987821864, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133870868641, "tokens_per_second": 153707.61262290328, "data_wait_s": 9.931998647516593e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 914700, "loss": 0.6132705807685852, "lr": 3e-05, "grad_norm": 0.36856665212879103, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133885605938, "tokens_per_second": 154593.4702946775, "data_wait_s": 1.4293000276666135e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 914800, "loss": 0.5447489023208618, "lr": 3e-05, "grad_norm": 0.34278712036988007, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133900311472, "tokens_per_second": 126826.12141213288, "data_wait_s": 1.4359000488184392e-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 sha256:7bd4890d43bce5a81f02622edf3c825d94ba0a4127d9cbac5446fc5e4a2c1e98
-size 633126
+oid sha256:77079a73d53ab6f8767e984716a380935c765efd3277c8a8ffe23ee9c4dabd01
+size 635244