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": 76014.04676912409,
-      "required_pages_per_second_upper_estimate": 0.006588704918107485
+      "required_token_positions_per_second": 74567.37030976762,
+      "required_pages_per_second_upper_estimate": 0.006463310669178431
     },
     {
       "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": 9501.755846140511,
-      "required_pages_per_second_upper_estimate": 0.0037903924709352605
+      "required_token_positions_per_second": 9320.921288720952,
+      "required_pages_per_second_upper_estimate": 0.0037182548622630256
     },
     {
       "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": 19003.511692281023,
-      "required_pages_per_second_upper_estimate": 0.004396127979626329
+      "required_token_positions_per_second": 18641.842577441905,
+      "required_pages_per_second_upper_estimate": 0.004312462195067324
     },
     {
       "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": 28505.267538421533,
-      "required_pages_per_second_upper_estimate": 0.0017358212375093616
+      "required_token_positions_per_second": 27962.763866162855,
+      "required_pages_per_second_upper_estimate": 0.001702785610165603
     },
     {
       "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": 28505.267538421533,
-      "required_pages_per_second_upper_estimate": 0.002194414402738549
+      "required_token_positions_per_second": 27962.763866162855,
+      "required_pages_per_second_upper_estimate": 0.0021526509683017954
     },
     {
       "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": 28505.267538421533,
-      "required_pages_per_second_upper_estimate": 0.35394002183371037
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27962.763866162855,
+      "required_pages_per_second_upper_estimate": 0.3472039418672518
     }
   ],
   "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": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09376374821076183,
-    "producer_p95_seconds": 0.3290863529982744,
-    "gpu_update_seconds": 0.9699259957035645,
-    "producer_batches_per_second": 10.66510265515627,
-    "gpu_batches_per_second": 1.0310064937218435,
-    "producer_headroom": 10.344360312083175,
+    "producer_batch_seconds": 0.07142579928111559,
+    "producer_p95_seconds": 0.282843322998815,
+    "gpu_update_seconds": 0.9887434637123355,
+    "producer_batches_per_second": 14.000543361989259,
+    "gpu_batches_per_second": 1.0113846884462838,
+    "producer_headroom": 13.842945737588005,
     "sustainable": true,
-    "buffer_seconds": 60.135411733621,
+    "buffer_seconds": 60.313351286452466,
     "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": 505013.24698927026,
-    "upload_bytes_per_second": 5704924.515516374,
+    "generation_bytes_per_second": 505608.58266531146,
+    "upload_bytes_per_second": 5620699.6000350425,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 108.36408669718342,
+    "estimated_full_shards_per_hour": 108.49183187455662,
     "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": 0.9722155079998629,
-      "bytes": 6032233,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.1764326539996546,
       "bytes": 2559639,
@@ -562,6 +556,12 @@
       "bytes": 6235096,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.1694963700028893,
+      "bytes": 895144,
+      "pages": 30,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -197,3 +197,4 @@
 {"step": 134700, "loss": 0.6082310862839222, "lr": 0.0003, "grad_norm": 0.1380898777569113, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19712934063, "tokens_per_second": 148904.92096440226, "data_wait_s": 1.4512999769067392e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 134800, "loss": 0.6221358515322208, "lr": 0.0003, "grad_norm": 0.16071592332279125, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19727550637, "tokens_per_second": 148767.60806423056, "data_wait_s": 1.568999869050458e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 134900, "loss": 0.6453112438321114, "lr": 0.0003, "grad_norm": 0.1575750167411725, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19742319510, "tokens_per_second": 150326.26741005166, "data_wait_s": 8.969000191427767e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 135000, "loss": 0.6044514365494251, "lr": 0.0003, "grad_norm": 0.12881867767854382, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19757015197, "tokens_per_second": 149972.18968166088, "data_wait_s": 8.879997039912269e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2b5aa7710976842e000838bfa9bf45e3e2c0f0a10ed5b59a85faf3a95c90d57b
-size 423996
+oid sha256:f74507f726c663b30934d322dd22c369a4b88ad3f368623d1981ab71ba738c5d
+size 426114