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": 75629.65283278764,
-      "required_pages_per_second_upper_estimate": 0.0065553866259433664
+      "required_token_positions_per_second": 73561.44211015785,
+      "required_pages_per_second_upper_estimate": 0.006376119362337969
     },
     {
       "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": 9453.706604098456,
-      "required_pages_per_second_upper_estimate": 0.0037712249098844964
+      "required_token_positions_per_second": 9195.180263769731,
+      "required_pages_per_second_upper_estimate": 0.003668094887414126
     },
     {
       "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": 18907.41320819691,
-      "required_pages_per_second_upper_estimate": 0.004373897286609025
+      "required_token_positions_per_second": 18390.360527539462,
+      "required_pages_per_second_upper_estimate": 0.0042542862487553075
     },
     {
       "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": 28361.119812295365,
-      "required_pages_per_second_upper_estimate": 0.0017270433972729482
+      "required_token_positions_per_second": 27585.540791309195,
+      "required_pages_per_second_upper_estimate": 0.001679814704043533
     },
     {
       "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": 28361.119812295365,
-      "required_pages_per_second_upper_estimate": 0.002183317511754912
+      "required_token_positions_per_second": 27585.540791309195,
+      "required_pages_per_second_upper_estimate": 0.002123611291779256
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28361.119812295365,
-      "required_pages_per_second_upper_estimate": 0.3521501895066288
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27585.540791309195,
+      "required_pages_per_second_upper_estimate": 0.3425200937619876
     }
   ],
   "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": 62,
+    "prefetch_batches": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.099067899195461,
-    "producer_p95_seconds": 0.37787702799687395,
-    "gpu_update_seconds": 0.9748557244207894,
-    "producer_batches_per_second": 10.094087066760139,
-    "gpu_batches_per_second": 1.0257928172849886,
-    "producer_headroom": 9.840278559832976,
+    "producer_batch_seconds": 0.21078979967171563,
+    "producer_p95_seconds": 0.3949673250026535,
+    "gpu_update_seconds": 1.002264200987152,
+    "producer_batches_per_second": 4.744062575880814,
+    "gpu_batches_per_second": 0.9977409140375142,
+    "producer_headroom": 4.754804087048234,
     "sustainable": true,
-    "buffer_seconds": 60.441054914088944,
+    "buffer_seconds": 60.13585205922912,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 380869.328477948,
-    "upload_bytes_per_second": 5263316.415894249,
+    "generation_bytes_per_second": 391704.6492992387,
+    "upload_bytes_per_second": 5398078.454675434,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.72569170717077,
+    "estimated_full_shards_per_hour": 84.05069932206031,
     "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": 0.8494216189974395,
-      "bytes": 1181935,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 3.5161270839998906,
       "bytes": 6159420,
@@ -562,6 +556,12 @@
       "bytes": 924946,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.1823451020027278,
+      "bytes": 13453015,
+      "pages": 42,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -381,3 +381,4 @@
 {"step": 393100, "loss": 0.6420015506446362, "lr": 0.0003, "grad_norm": 0.1850644812418594, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57539130193, "tokens_per_second": 149961.80861544103, "data_wait_s": 9.725001291371882e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 393200, "loss": 0.6698104962706566, "lr": 0.0003, "grad_norm": 0.18173146493014478, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57553791015, "tokens_per_second": 149230.4514493089, "data_wait_s": 9.944997145794332e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 393300, "loss": 0.5766168162226677, "lr": 0.0003, "grad_norm": 0.16535435980387686, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57568269703, "tokens_per_second": 146431.22659633058, "data_wait_s": 1.585899735800922e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 393400, "loss": 0.5997443683445454, "lr": 0.0003, "grad_norm": 0.20393893049244358, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57582797421, "tokens_per_second": 149788.42920753785, "data_wait_s": 1.2819997209589928e-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 sha256:7ac190699c8bfcb1bdffe7c413f8e1fc5e83f1e5cd71f55c0a1c0c88972ac87f
-size 816840
+oid sha256:469b90e24bfc5d90e483ff46aada6565c695b2e03595a6bd5dc58a2c3b851a8d
+size 818958