Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74912.62288844607,
-      "required_pages_per_second_upper_estimate": 0.006493236287663619
+      "required_token_positions_per_second": 75325.43003370069,
+      "required_pages_per_second_upper_estimate": 0.006529017364764144
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9364.077861055759,
-      "required_pages_per_second_upper_estimate": 0.003735470664215637
+      "required_token_positions_per_second": 9415.678754212586,
+      "required_pages_per_second_upper_estimate": 0.0037560550319979996
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18728.155722111518,
-      "required_pages_per_second_upper_estimate": 0.004332429222027175
+      "required_token_positions_per_second": 18831.35750842517,
+      "required_pages_per_second_upper_estimate": 0.004356303139001433
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28092.23358316728,
-      "required_pages_per_second_upper_estimate": 0.0017106696366560644
+      "required_token_positions_per_second": 28247.036262637754,
+      "required_pages_per_second_upper_estimate": 0.0017200963076489263
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28092.23358316728,
-      "required_pages_per_second_upper_estimate": 0.0021626179055119106
+      "required_token_positions_per_second": 28247.036262637754,
+      "required_pages_per_second_upper_estimate": 0.002174535044298791
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 28092.23358316728,
-      "required_pages_per_second_upper_estimate": 0.3488115224451777
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28247.036262637754,
+      "required_pages_per_second_upper_estimate": 0.35073365363296066
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.11897584066392142,
-    "producer_p95_seconds": 0.3253241229976993,
-    "gpu_update_seconds": 0.9841866051037871,
-    "producer_batches_per_second": 8.405067738287837,
-    "gpu_batches_per_second": 1.0160674762430295,
-    "producer_headroom": 8.272155083012871,
+    "producer_batch_seconds": 0.0728630952891649,
+    "producer_p95_seconds": 0.28476002599927597,
+    "gpu_update_seconds": 0.9787929516899407,
+    "producer_batches_per_second": 13.724368914488112,
+    "gpu_batches_per_second": 1.0216665314900808,
+    "producer_headroom": 13.433315559893487,
     "sustainable": true,
-    "buffer_seconds": 60.03538291133101,
+    "buffer_seconds": 60.68516300477633,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 283818.7216043086,
+    "generation_bytes_per_second": 321336.6348154122,
     "upload_bytes_per_second": 5510506.701707024,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.90089069458907,
+    "estimated_full_shards_per_hour": 68.95136149737144,
     "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-v1/1790782192342215557-281f6532/train.jsonl

@@ -227,3 +227,4 @@
 {"step": 377700, "loss": 0.627464709803462, "lr": 0.0003, "grad_norm": 0.15065318497035127, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55284306130, "tokens_per_second": 150282.58436663955, "data_wait_s": 1.548100044601597e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 377800, "loss": 0.6744170840829611, "lr": 0.0003, "grad_norm": 0.16375207792487784, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55298975790, "tokens_per_second": 149346.32428860036, "data_wait_s": 1.090099976863712e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 377900, "loss": 0.6095782984048128, "lr": 0.0003, "grad_norm": 0.15612479376576685, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55313733563, "tokens_per_second": 150793.86810920294, "data_wait_s": 9.723000403027982e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 378000, "loss": 0.6229256056249142, "lr": 0.0003, "grad_norm": 0.18372337081710355, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55328384108, "tokens_per_second": 148847.05696519098, "data_wait_s": 1.2398999388096854e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0801e6dc960d3e8f91ed452be658ebcb7ca5a50d920d7b370aa80670e3596a3b
-size 488319
+oid sha256:b70aa542e100e726f646f6ad1efba3a0db37bd88686a8bd1fdf09e881c155568
+size 490437