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": 75838.56606245865,
-      "required_pages_per_second_upper_estimate": 0.0065734946952055095
+      "required_token_positions_per_second": 74318.19506116002,
+      "required_pages_per_second_upper_estimate": 0.0064417127901580185
     },
     {
       "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": 9479.82075780733,
-      "required_pages_per_second_upper_estimate": 0.0037816422362403584
+      "required_token_positions_per_second": 9289.774382645002,
+      "required_pages_per_second_upper_estimate": 0.0037058298957415837
     },
     {
       "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": 18959.64151561466,
-      "required_pages_per_second_upper_estimate": 0.004385979386343306
+      "required_token_positions_per_second": 18579.548765290005,
+      "required_pages_per_second_upper_estimate": 0.004298051617959631
     },
     {
       "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": 28439.462273421992,
-      "required_pages_per_second_upper_estimate": 0.0017318140421244323
+      "required_token_positions_per_second": 27869.323147935007,
+      "required_pages_per_second_upper_estimate": 0.0016970955606710875
     },
     {
       "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": 28439.462273421992,
-      "required_pages_per_second_upper_estimate": 0.002189348531278252
+      "required_token_positions_per_second": 27869.323147935007,
+      "required_pages_per_second_upper_estimate": 0.0021454576431521525
     },
     {
       "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": 28439.462273421992,
-      "required_pages_per_second_upper_estimate": 0.35312294067847066
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27869.323147935007,
+      "required_pages_per_second_upper_estimate": 0.3460437208728287
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.2708600023828396,
-    "producer_p95_seconds": 0.18629149499975028,
-    "gpu_update_seconds": 0.9721702799506989,
-    "producer_batches_per_second": 3.6919441453248516,
-    "gpu_batches_per_second": 1.0286263843106913,
-    "producer_headroom": 3.5891983733228052,
+    "producer_batch_seconds": 0.06076905342968786,
+    "producer_p95_seconds": 0.18966419600019435,
+    "gpu_update_seconds": 0.9920585388184641,
+    "producer_batches_per_second": 16.455744224435527,
+    "gpu_batches_per_second": 1.0080050328390844,
+    "producer_headroom": 16.32506157046389,
     "sustainable": true,
-    "buffer_seconds": 60.27455735694333,
+    "buffer_seconds": 60.51557086792631,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 385864.63267285033,
-    "upload_bytes_per_second": 6145009.125047967,
+    "generation_bytes_per_second": 371043.02612122666,
+    "upload_bytes_per_second": 6219993.36979219,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.7975677026666,
+    "estimated_full_shards_per_hour": 79.61719596603012,
     "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": [
@@ -358,6 +358,12 @@
       "bytes": 2397832,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.344766387999698,
+      "bytes": 10673063,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -61,3 +61,4 @@
 {"step": 46100, "loss": 0.6462961342185736, "lr": 0.0003, "grad_norm": 0.12934043426875, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6745887842, "tokens_per_second": 148537.8203293296, "data_wait_s": 1.0992000170517713e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 46200, "loss": 0.6512833032757044, "lr": 0.0003, "grad_norm": 0.14297691303761664, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6760451345, "tokens_per_second": 148569.01473379356, "data_wait_s": 1.332300053036306e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 46300, "loss": 0.6293332725763321, "lr": 0.0003, "grad_norm": 0.14301501869759042, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6774948125, "tokens_per_second": 147025.47778413142, "data_wait_s": 1.3222999768913724e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 46400, "loss": 0.6492524016648531, "lr": 0.0003, "grad_norm": 0.14303506288905476, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6789668132, "tokens_per_second": 150701.92041407048, "data_wait_s": 1.6092999430838972e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c1efa2e5d3807059a33fac2a9de1b3d633bff74a04fe24a615c2491e817366bb
-size 134382
+oid sha256:fb0421f1ac61445010b525a00ac932a2c9f28270659299f2df0ef6c1cf928b82
+size 136500