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": 76141.60687187794,
-      "required_pages_per_second_upper_estimate": 0.00526765303626523
+      "required_token_positions_per_second": 75827.67668167078,
+      "required_pages_per_second_upper_estimate": 0.005245934617288301
     },
     {
       "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": 9517.700858984743,
-      "required_pages_per_second_upper_estimate": 0.003984862597462111
+      "required_token_positions_per_second": 9478.459585208848,
+      "required_pages_per_second_upper_estimate": 0.0039684330955831225
     },
     {
       "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": 19035.401717969486,
-      "required_pages_per_second_upper_estimate": 0.003964572180551885
+      "required_token_positions_per_second": 18956.919170417696,
+      "required_pages_per_second_upper_estimate": 0.003948226335620809
     },
     {
       "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": 28553.10257695423,
-      "required_pages_per_second_upper_estimate": 0.0017117149014251808
+      "required_token_positions_per_second": 28435.378755626545,
+      "required_pages_per_second_upper_estimate": 0.001704657538090453
     },
     {
       "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": 28553.10257695423,
-      "required_pages_per_second_upper_estimate": 0.0022198781335329085
+      "required_token_positions_per_second": 28435.378755626545,
+      "required_pages_per_second_upper_estimate": 0.0022107256242370472
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28553.10257695423,
-      "required_pages_per_second_upper_estimate": 0.36624854192422146
+      "required_token_positions_per_second": 28435.378755626545,
+      "required_pages_per_second_upper_estimate": 0.36473850714622114
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.22824270127343027,
-    "producer_p95_seconds": 0.22941520899985335,
-    "gpu_update_seconds": 0.9683010778070488,
-    "producer_batches_per_second": 4.381301108078017,
-    "gpu_batches_per_second": 1.0327366383446985,
-    "producer_headroom": 4.242418585149162,
+    "producer_batch_seconds": 0.28890136149999535,
+    "producer_p95_seconds": 0.22820466600023792,
+    "gpu_update_seconds": 0.9723098903519706,
+    "producer_batches_per_second": 3.4613890180646174,
+    "gpu_batches_per_second": 1.0284786876311685,
+    "producer_headroom": 3.3655427766199235,
     "sustainable": true,
-    "buffer_seconds": 60.034666824037025,
+    "buffer_seconds": 60.28321320182218,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 491050.5838319743,
-    "upload_bytes_per_second": 5112430.952778343,
+    "generation_bytes_per_second": 468016.55123318936,
+    "upload_bytes_per_second": 5025704.432303567,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.36802421779082,
+    "estimated_full_shards_per_hour": 100.42545702692757,
     "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": 1.1100294799998665,
-      "bytes": 10655210,
-      "pages": 43,
-      "shards": 3
-    },
     {
       "seconds": 0.9483619619995807,
       "bytes": 1257250,
@@ -562,6 +556,12 @@
       "bytes": 7505854,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.4834884449992387,
+      "bytes": 5324170,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -151,3 +151,4 @@
 {"step": 455300, "loss": 0.58512943983078, "lr": 0.0003, "grad_norm": 0.18817986255199826, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66642029194, "tokens_per_second": 153155.42990572797, "data_wait_s": 1.2401998901623301e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 455400, "loss": 0.6065024137496948, "lr": 0.0003, "grad_norm": 0.1634300136332109, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66656557049, "tokens_per_second": 152039.80700311853, "data_wait_s": 1.4707000445923768e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 455500, "loss": 0.6034132242202759, "lr": 0.0003, "grad_norm": 0.1672609640580655, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66671161434, "tokens_per_second": 150856.0205691735, "data_wait_s": 1.2145999789936468e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 455600, "loss": 0.5722441077232361, "lr": 0.0003, "grad_norm": 0.13953617455733194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66685849006, "tokens_per_second": 152365.03168072176, "data_wait_s": 1.671499921940267e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5870d2a62b460d8dd8597a581d166b06e392646d5819a95a9b437e3aab62ea7b
-size 326568
+oid sha256:892e65822f838674af8f099392d958509c6ae5c6ac6bd9814bbfd46bfdf5108e
+size 328686