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": 75464.72918787059,
-      "required_pages_per_second_upper_estimate": 0.005220825067512992
+      "required_token_positions_per_second": 76760.39387079033,
+      "required_pages_per_second_upper_estimate": 0.0053104621566336385
     },
     {
       "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": 9433.091148483823,
-      "required_pages_per_second_upper_estimate": 0.0039494382785270465
+      "required_token_positions_per_second": 9595.049233848791,
+      "required_pages_per_second_upper_estimate": 0.004017246746799948
     },
     {
       "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": 18866.182296967647,
-      "required_pages_per_second_upper_estimate": 0.003929328237773431
+      "required_token_positions_per_second": 19190.098467697582,
+      "required_pages_per_second_upper_estimate": 0.003996791433892614
     },
     {
       "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": 28299.27344545147,
-      "required_pages_per_second_upper_estimate": 0.0016964982325662328
+      "required_token_positions_per_second": 28785.147701546375,
+      "required_pages_per_second_upper_estimate": 0.00172562565233209
     },
     {
       "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": 28299.27344545147,
-      "required_pages_per_second_upper_estimate": 0.0022001440350349257
+      "required_token_positions_per_second": 28785.147701546375,
+      "required_pages_per_second_upper_estimate": 0.002237918621311311
     },
     {
       "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": 28299.27344545147,
-      "required_pages_per_second_upper_estimate": 0.36299269436579146
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28785.147701546375,
+      "required_pages_per_second_upper_estimate": 0.36922496763184637
     }
   ],
   "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": 0.326987140921851,
-    "producer_p95_seconds": 0.1670731189997241,
-    "gpu_update_seconds": 0.9769862132076699,
-    "producer_batches_per_second": 3.058224238362319,
-    "gpu_batches_per_second": 1.0235558971879148,
-    "producer_headroom": 2.9878429177775123,
+    "producer_batch_seconds": 0.049377420015630946,
+    "producer_p95_seconds": 0.16342882300068595,
+    "gpu_update_seconds": 0.9604953320602456,
+    "producer_batches_per_second": 20.252171937769113,
+    "gpu_batches_per_second": 1.04112947415894,
+    "producer_headroom": 19.45211661030873,
     "sustainable": true,
-    "buffer_seconds": 60.57314521887553,
+    "buffer_seconds": 60.51120591979547,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 320849.2077964165,
-    "upload_bytes_per_second": 5939941.583462157,
+    "generation_bytes_per_second": 360964.519445533,
+    "upload_bytes_per_second": 5956950.917595794,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.84677100581523,
+    "estimated_full_shards_per_hour": 77.45458304905408,
     "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": [
@@ -346,6 +346,12 @@
       "bytes": 2583341,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.5200459669995325,
+      "bytes": 9596109,
+      "pages": 30,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -38,3 +38,4 @@
 {"step": 488800, "loss": 0.5863213539123535, "lr": 0.0003, "grad_norm": 0.15694707538758185, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71544182605, "tokens_per_second": 153333.76972484827, "data_wait_s": 1.2375000551401172e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 488900, "loss": 0.6243778467178345, "lr": 0.0003, "grad_norm": 0.18731338743514026, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71558886946, "tokens_per_second": 152825.46945399616, "data_wait_s": 7.768000614305492e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 489000, "loss": 0.589903712272644, "lr": 0.0003, "grad_norm": 0.17181293917392576, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71573585497, "tokens_per_second": 153517.7667324121, "data_wait_s": 1.5919999896141235e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 489100, "loss": 0.5786427855491638, "lr": 0.0003, "grad_norm": 0.1550712551863727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71588238509, "tokens_per_second": 153157.35521647448, "data_wait_s": 1.296700065722689e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:817f34fc730d0259cc07a6b6231df5d66071474768f6124ce88ee220442447d0
-size 84885
+oid sha256:639e03de6ca59384e41f3b73ee4a13acc420ebf43231bf8a04da526d6d293c3d
+size 87003