Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76645.00759421651,
-      "required_pages_per_second_upper_estimate": 0.005302479466287214
+      "required_token_positions_per_second": 77144.5045607611,
+      "required_pages_per_second_upper_estimate": 0.005337035825425409
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9580.625949277064,
-      "required_pages_per_second_upper_estimate": 0.0040112080187421975
+      "required_token_positions_per_second": 9643.063070095137,
+      "required_pages_per_second_upper_estimate": 0.004037349137393378
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19161.25189855413,
-      "required_pages_per_second_upper_estimate": 0.003990783454275218
+      "required_token_positions_per_second": 19286.126140190274,
+      "required_pages_per_second_upper_estimate": 0.00401679146565282
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28741.877847831194,
-      "required_pages_per_second_upper_estimate": 0.0017230316906711061
+      "required_token_positions_per_second": 28929.18921028541,
+      "required_pages_per_second_upper_estimate": 0.0017342607208423462
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28741.877847831194,
-      "required_pages_per_second_upper_estimate": 0.0022345545805089284
+      "required_token_positions_per_second": 28929.18921028541,
+      "required_pages_per_second_upper_estimate": 0.0022491172150441324
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28741.877847831194,
-      "required_pages_per_second_upper_estimate": 0.3686699484079372
+      "required_token_positions_per_second": 28929.18921028541,
+      "required_pages_per_second_upper_estimate": 0.3710725774462284
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11709323212505751,
-    "producer_p95_seconds": 0.34080561099835904,
-    "gpu_update_seconds": 0.9619413229148583,
-    "producer_batches_per_second": 8.540203236784714,
-    "gpu_batches_per_second": 1.0395644476212094,
-    "producer_headroom": 8.215174399554442,
+    "producer_batch_seconds": 0.09812232820308964,
+    "producer_p95_seconds": 0.34375213400198845,
+    "gpu_update_seconds": 0.9557129236850544,
+    "producer_batches_per_second": 10.191360298037775,
+    "gpu_batches_per_second": 1.0463393088210868,
+    "producer_headroom": 9.74001474676547,
     "sustainable": true,
-    "buffer_seconds": 60.60230334363607,
+    "buffer_seconds": 60.20991419215843,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 406779.58029271016,
+    "generation_bytes_per_second": 406618.36189258524,
     "upload_bytes_per_second": 5126782.612468387,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.28542858682611,
+    "estimated_full_shards_per_hour": 87.25083487113159,
     "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-v1/1790824801708454555-41033672/train.jsonl

@@ -387,3 +387,4 @@
 {"step": 435800, "loss": 0.9287593364715576, "lr": 0.0003, "grad_norm": 1.9048118937577978, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63788704984, "tokens_per_second": 152400.58610558434, "data_wait_s": 1.6085999959614128e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 435900, "loss": 0.5809106826782227, "lr": 0.0003, "grad_norm": 0.2027705148394604, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63803264435, "tokens_per_second": 151935.50790462716, "data_wait_s": 1.3031996786594391e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 436000, "loss": 0.4993954300880432, "lr": 0.0003, "grad_norm": 0.14581099757722213, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63817751974, "tokens_per_second": 150349.26657984426, "data_wait_s": 1.1990996426902711e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 436100, "loss": 0.5766322016716003, "lr": 0.0003, "grad_norm": 0.15279914796720345, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63832285536, "tokens_per_second": 151045.89412842927, "data_wait_s": 2.7783004043158144e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5646ab3b8fa9a16c49b65140922ea7d5497b9998d888b48f9e8416baf5e493b3
-size 830331
+oid sha256:d9457fe292a7a998f1c2a67d262424e52617ef865c186144d79b7a64cfd603d2
+size 832449