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": 74789.45355307494,
-      "required_pages_per_second_upper_estimate": 0.006482560281843476
+      "required_token_positions_per_second": 75601.86866191775,
+      "required_pages_per_second_upper_estimate": 0.006552978364430701
     },
     {
       "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": 9348.681694134368,
-      "required_pages_per_second_upper_estimate": 0.003729328903037485
+      "required_token_positions_per_second": 9450.233582739718,
+      "required_pages_per_second_upper_estimate": 0.0037698394697381994
     },
     {
       "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": 18697.363388268735,
-      "required_pages_per_second_upper_estimate": 0.004325305957519213
+      "required_token_positions_per_second": 18900.467165479437,
+      "required_pages_per_second_upper_estimate": 0.004372290441872514
     },
     {
       "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": 28046.045082403103,
-      "required_pages_per_second_upper_estimate": 0.0017078569992918613
+      "required_token_positions_per_second": 28350.70074821916,
+      "required_pages_per_second_upper_estimate": 0.0017264089309353117
     },
     {
       "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": 28046.045082403103,
-      "required_pages_per_second_upper_estimate": 0.0021590621868651312
+      "required_token_positions_per_second": 28350.70074821916,
+      "required_pages_per_second_upper_estimate": 0.00218251542336051
     },
     {
       "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": 28046.045082403103,
-      "required_pages_per_second_upper_estimate": 0.3482380158486547
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28350.70074821916,
+      "required_pages_per_second_upper_estimate": 0.35202081960116666
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.0880060003749179,
-    "producer_p95_seconds": 0.3666138389999105,
-    "gpu_update_seconds": 0.9858074433941187,
-    "producer_batches_per_second": 11.36286157466377,
-    "gpu_batches_per_second": 1.0143968852142327,
-    "producer_headroom": 11.201593518560562,
+    "producer_batch_seconds": 0.1115278211328814,
+    "producer_p95_seconds": 0.3788880829997652,
+    "gpu_update_seconds": 0.9752139901422614,
+    "producer_batches_per_second": 8.966372604092532,
+    "gpu_batches_per_second": 1.025415970349362,
+    "producer_headroom": 8.744132004339338,
     "sustainable": true,
-    "buffer_seconds": 60.134254047041246,
+    "buffer_seconds": 60.463267388820206,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 336388.06727964967,
-    "upload_bytes_per_second": 6362207.785181299,
+    "generation_bytes_per_second": 390148.1906052726,
+    "upload_bytes_per_second": 6240443.286975443,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.18104852478139,
+    "estimated_full_shards_per_hour": 83.7167195188392,
     "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": 0.9525506049976684,
-      "bytes": 10860039,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.2853242279998085,
       "bytes": 7445426,
@@ -562,6 +556,12 @@
       "bytes": 5863488,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.6997049530000368,
+      "bytes": 1148192,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -311,3 +311,4 @@
 {"step": 31100, "loss": 0.6702503003180027, "lr": 0.0003, "grad_norm": 0.13519765288437588, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4551005576, "tokens_per_second": 149149.51730247418, "data_wait_s": 1.0536998161114752e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 31200, "loss": 0.664759548380971, "lr": 0.0003, "grad_norm": 0.13522052249851715, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4565619590, "tokens_per_second": 149727.55217065001, "data_wait_s": 8.682000043336302e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 31300, "loss": 0.6205551885068417, "lr": 0.0003, "grad_norm": 0.12686401793089822, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4580321298, "tokens_per_second": 148694.87510512723, "data_wait_s": 1.2466000043787062e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 31400, "loss": 0.6879486478865147, "lr": 0.0003, "grad_norm": 0.16777613929339316, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4595072175, "tokens_per_second": 150432.5182586362, "data_wait_s": 1.0510000720387325e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0822ce8c5a5449d2a6c53d0bb3db281209bd6ffd502f724e79924e89dce030d1
-size 662023
+oid sha256:dc4e36d9a26038571493277cba2f40229681886b245149b0c553fb83b6e8c0a5
+size 664141