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": 75123.69833907967,
-      "required_pages_per_second_upper_estimate": 0.0065115317727587
+      "required_token_positions_per_second": 74242.25494401567,
+      "required_pages_per_second_upper_estimate": 0.006435130493272413
     },
     {
       "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": 9390.462292384958,
-      "required_pages_per_second_upper_estimate": 0.003745995808355257
+      "required_token_positions_per_second": 9280.281868001959,
+      "required_pages_per_second_upper_estimate": 0.0037020431897247324
     },
     {
       "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": 18780.924584769917,
-      "required_pages_per_second_upper_estimate": 0.004344636369702931
+      "required_token_positions_per_second": 18560.563736003918,
+      "required_pages_per_second_upper_estimate": 0.004293659765559398
     },
     {
       "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": 28171.386877154877,
-      "required_pages_per_second_upper_estimate": 0.0017154896569746706
+      "required_token_positions_per_second": 27840.845604005877,
+      "required_pages_per_second_upper_estimate": 0.0016953614276559294
     },
     {
       "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": 28171.386877154877,
-      "required_pages_per_second_upper_estimate": 0.0021687113451934175
+      "required_token_positions_per_second": 27840.845604005877,
+      "required_pages_per_second_upper_estimate": 0.002143265362989601
     },
     {
       "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": 28171.386877154877,
-      "required_pages_per_second_upper_estimate": 0.3497943414474698
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27840.845604005877,
+      "required_pages_per_second_upper_estimate": 0.34569012508543745
     }
   ],
   "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": 0.09110067318729875,
-    "producer_p95_seconds": 0.2556450119991496,
-    "gpu_update_seconds": 0.981421330819204,
-    "producer_batches_per_second": 10.976867294317863,
-    "gpu_batches_per_second": 1.0189303702674652,
-    "producer_headroom": 10.772931708215232,
+    "producer_batch_seconds": 0.11990261614835163,
+    "producer_p95_seconds": 0.2982966640011,
+    "gpu_update_seconds": 0.9930732849587683,
+    "producer_batches_per_second": 8.340101593469257,
+    "gpu_batches_per_second": 1.0069750290800736,
+    "producer_headroom": 8.282332086316373,
     "sustainable": true,
-    "buffer_seconds": 60.84812251079065,
+    "buffer_seconds": 60.57747038248487,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 368753.8941334714,
+    "generation_bytes_per_second": 419162.81778057065,
     "upload_bytes_per_second": 5515677.652446132,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.12600152972324,
+    "estimated_full_shards_per_hour": 89.94258308470573,
     "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": [

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -210,3 +210,4 @@
 {"step": 336000, "loss": 0.6260675601661205, "lr": 0.0003, "grad_norm": 0.15267251646496335, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49181851800, "tokens_per_second": 149769.7196090554, "data_wait_s": 8.497001545038074e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 336100, "loss": 0.6162252761423588, "lr": 0.0003, "grad_norm": 0.15455633509498237, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49196560391, "tokens_per_second": 149375.19740666906, "data_wait_s": 1.2152999261161312e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 336200, "loss": 0.5848078187555075, "lr": 0.0003, "grad_norm": 0.14169624493234997, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49211216171, "tokens_per_second": 148733.23294222055, "data_wait_s": 1.329800215899013e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 336300, "loss": 0.5717901196330786, "lr": 0.0003, "grad_norm": 0.15387478381925662, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49225815047, "tokens_per_second": 147822.41502582183, "data_wait_s": 1.6064997907960787e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6fb01ddce7a7181307d1dd12b10e07fd8f1c791d882602ef6c9ffaf2cfd379e5
-size 452313
+oid sha256:324bf31688de86afc6db04ff9502bbb5b99326279b4efee806765664a5831e84
+size 454431