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": 76217.14963508995,
-      "required_pages_per_second_upper_estimate": 0.006606309359769898
+      "required_token_positions_per_second": 75789.62518306063,
+      "required_pages_per_second_upper_estimate": 0.0065692526238188146
     },
     {
       "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": 9527.143704386244,
-      "required_pages_per_second_upper_estimate": 0.0038005200671717906
+      "required_token_positions_per_second": 9473.70314788258,
+      "required_pages_per_second_upper_estimate": 0.003779201830174956
     },
     {
       "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": 19054.28740877249,
-      "required_pages_per_second_upper_estimate": 0.004407874048014541
+      "required_token_positions_per_second": 18947.40629576516,
+      "required_pages_per_second_upper_estimate": 0.004383148983563638
     },
     {
       "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": 28581.43111315873,
-      "required_pages_per_second_upper_estimate": 0.001740459199611463
+      "required_token_positions_per_second": 28421.109443647736,
+      "required_pages_per_second_upper_estimate": 0.0017306964510810363
     },
     {
       "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": 28581.43111315873,
-      "required_pages_per_second_upper_estimate": 0.002200277685556087
+      "required_token_positions_per_second": 28421.109443647736,
+      "required_pages_per_second_upper_estimate": 0.0021879356796383454
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28581.43111315873,
-      "required_pages_per_second_upper_estimate": 0.3548857185288592
+      "required_token_positions_per_second": 28421.109443647736,
+      "required_pages_per_second_upper_estimate": 0.35289505995564446
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08641562344541853,
-    "producer_p95_seconds": 0.24986374199943384,
-    "gpu_update_seconds": 0.9673413444742106,
-    "producer_batches_per_second": 11.571981548355266,
-    "gpu_batches_per_second": 1.0337612526460767,
-    "producer_headroom": 11.19405618921674,
+    "producer_batch_seconds": 0.38576812950788053,
+    "producer_p95_seconds": 0.3160173420001229,
+    "gpu_update_seconds": 0.97279805543198,
+    "producer_batches_per_second": 2.5922307300908636,
+    "gpu_batches_per_second": 1.0279625811504534,
+    "producer_headroom": 2.5217170134634137,
     "sustainable": true,
-    "buffer_seconds": 60.942504701875265,
+    "buffer_seconds": 60.31347943678276,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 342093.8749586195,
-    "upload_bytes_per_second": 5083369.4816179685,
+    "generation_bytes_per_second": 295931.81346536905,
+    "upload_bytes_per_second": 5238317.9402646115,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.40538202828348,
+    "estimated_full_shards_per_hour": 63.500078229625736,
     "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": [
-    {
-      "seconds": 0.9314350770000601,
-      "bytes": 2320145,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.1146210560000327,
       "bytes": 5994268,
@@ -562,6 +556,12 @@
       "bytes": 914174,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.4923641690002114,
+      "bytes": 17877863,
+      "pages": 40,
+      "shards": 5
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -175,3 +175,4 @@
 {"step": 172500, "loss": 0.5745960976928473, "lr": 0.0003, "grad_norm": 0.1404494812128547, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25246218209, "tokens_per_second": 150434.5973109403, "data_wait_s": 1.2188000255264342e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 172600, "loss": 0.6041794866323471, "lr": 0.0003, "grad_norm": 0.15712818106169318, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25260970011, "tokens_per_second": 151294.78968345665, "data_wait_s": 1.2436998076736927e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 172700, "loss": 0.5568112526088953, "lr": 0.0003, "grad_norm": 0.14181230128600136, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25275678110, "tokens_per_second": 128728.2962534748, "data_wait_s": 1.0804000339703634e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 172800, "loss": 0.6099232845008373, "lr": 0.0003, "grad_norm": 0.14864510725580335, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25290431838, "tokens_per_second": 150122.77375398081, "data_wait_s": 1.3106997357681394e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:72b1639aad8a9984bf529db64bfbf488150e54d2d166c11f19a8c695c60f4766
-size 377400
+oid sha256:5380da3075e0edb27a92f319943b9ad0f3e6cc2325735bab38054a9209393d2c
+size 379518