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": 75708.09017882668,
-      "required_pages_per_second_upper_estimate": 0.006562185376300394
+      "required_token_positions_per_second": 74655.20444380539,
+      "required_pages_per_second_upper_estimate": 0.006470923909303247
     },
     {
       "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": 9463.511272353335,
-      "required_pages_per_second_upper_estimate": 0.0037751361386442217
+      "required_token_positions_per_second": 9331.900555475673,
+      "required_pages_per_second_upper_estimate": 0.00372263465592615
     },
     {
       "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": 18927.02254470667,
-      "required_pages_per_second_upper_estimate": 0.004378433561498023
+      "required_token_positions_per_second": 18663.801110951346,
+      "required_pages_per_second_upper_estimate": 0.0043175419153377425
     },
     {
       "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": 28390.533817060004,
-      "required_pages_per_second_upper_estimate": 0.0017288345558397584
+      "required_token_positions_per_second": 27995.70166642702,
+      "required_pages_per_second_upper_estimate": 0.001704791349390408
     },
     {
       "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": 28390.533817060004,
-      "required_pages_per_second_upper_estimate": 0.0021855818832648703
+      "required_token_positions_per_second": 27995.70166642702,
+      "required_pages_per_second_upper_estimate": 0.002155186611343797
     },
     {
       "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": 28390.533817060004,
-      "required_pages_per_second_upper_estimate": 0.3525154130034643
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27995.70166642702,
+      "required_pages_per_second_upper_estimate": 0.34761291911080644
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.31357948278120773,
-    "producer_p95_seconds": 0.18834892100039724,
-    "gpu_update_seconds": 0.9738457254152152,
-    "producer_batches_per_second": 3.188984148869603,
-    "gpu_batches_per_second": 1.026856691878617,
-    "producer_headroom": 3.105578581793541,
+    "producer_batch_seconds": 0.243545942335885,
+    "producer_p95_seconds": 0.18490180600019812,
+    "gpu_update_seconds": 0.9875801767510621,
+    "producer_batches_per_second": 4.106001481317458,
+    "gpu_batches_per_second": 1.012576015134079,
+    "producer_headroom": 4.055005668659618,
     "sustainable": true,
-    "buffer_seconds": 60.37843497574334,
+    "buffer_seconds": 60.24239078181479,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 422421.0882053992,
-    "upload_bytes_per_second": 4714265.580201424,
+    "generation_bytes_per_second": 409554.13888470724,
+    "upload_bytes_per_second": 2877880.923329706,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.64173206922038,
+    "estimated_full_shards_per_hour": 87.8807842722503,
     "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": [
@@ -400,6 +400,12 @@
       "bytes": 936913,
       "pages": 30,
       "shards": 1
+    },
+    {
+      "seconds": 33.908926936999706,
+      "bytes": 9910795,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -43,3 +43,4 @@
 {"step": 241400, "loss": 0.6207055300474167, "lr": 0.0003, "grad_norm": 0.13743057772118955, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35333684821, "tokens_per_second": 149884.36847653115, "data_wait_s": 1.9915999473596457e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 241500, "loss": 0.58491569198668, "lr": 0.0003, "grad_norm": 0.15787964599663515, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35348409349, "tokens_per_second": 149601.08783378193, "data_wait_s": 1.3154000043869019e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 241600, "loss": 0.6268334165215492, "lr": 0.0003, "grad_norm": 0.14220717144270645, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35363108475, "tokens_per_second": 148962.41087267562, "data_wait_s": 9.933000001183245e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 241700, "loss": 0.5771536789834499, "lr": 0.0003, "grad_norm": 0.14329444178759554, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35377628896, "tokens_per_second": 148390.99048776756, "data_wait_s": 9.821000276133418e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:af8cc4a393f39057b5ab1e63187cc90556bce713642d0bbeccb77215a995f628
-size 96258
+oid sha256:523199c25b35ed5165b26e273c2790cc504a36923995843056bcb485859843ad
+size 98376