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": 76313.40763852552,
-      "required_pages_per_second_upper_estimate": 0.005279538611934608
+      "required_token_positions_per_second": 75990.13664305958,
+      "required_pages_per_second_upper_estimate": 0.005257173974376223
     },
     {
       "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": 9539.17595481569,
-      "required_pages_per_second_upper_estimate": 0.003993853771635532
+      "required_token_positions_per_second": 9498.767080382448,
+      "required_pages_per_second_upper_estimate": 0.003976935419743587
     },
     {
       "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": 19078.35190963138,
-      "required_pages_per_second_upper_estimate": 0.003973517572802334
+      "required_token_positions_per_second": 18997.534160764895,
+      "required_pages_per_second_upper_estimate": 0.0039566853670208076
     },
     {
       "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": 28617.527864447067,
-      "required_pages_per_second_upper_estimate": 0.0017155770990386584
+      "required_token_positions_per_second": 28496.301241147343,
+      "required_pages_per_second_upper_estimate": 0.0017083097480741758
     },
     {
       "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": 28617.527864447067,
-      "required_pages_per_second_upper_estimate": 0.002224886916258585
+      "required_token_positions_per_second": 28496.301241147343,
+      "required_pages_per_second_upper_estimate": 0.002215462079516595
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28617.527864447067,
-      "required_pages_per_second_upper_estimate": 0.36707492033769534
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28496.301241147343,
+      "required_pages_per_second_upper_estimate": 0.36551995537701343
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.2723217197581107,
-    "producer_p95_seconds": 0.3694382709945785,
-    "gpu_update_seconds": 0.9661211873702215,
-    "producer_batches_per_second": 3.6721272210246334,
-    "gpu_batches_per_second": 1.0350668353749664,
-    "producer_headroom": 3.54771991095083,
+    "producer_batch_seconds": 0.11128420523459681,
+    "producer_p95_seconds": 0.36455414200463565,
+    "gpu_update_seconds": 0.9702311807427683,
+    "producer_batches_per_second": 8.986001184012707,
+    "gpu_batches_per_second": 1.0306821918817761,
+    "producer_headroom": 8.718498538920562,
     "sustainable": true,
-    "buffer_seconds": 60.86563480432395,
+    "buffer_seconds": 60.15433320605163,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 160,
+    "upload_pages_per_commit": 167,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 633057.2603302192,
+    "generation_bytes_per_second": 609252.8511667543,
     "upload_bytes_per_second": 3866569.867299548,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.91967562403646,
+    "estimated_full_shards_per_hour": 65.36573958993898,
     "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-v1/1791300814123050567-72be911d/train.jsonl

@@ -349,3 +349,4 @@
 {"step": 834900, "loss": 0.5117660164833069, "lr": 3e-05, "grad_norm": 0.2857102331021943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122205454169, "tokens_per_second": 153142.0329584622, "data_wait_s": 2.237500302726403e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 835000, "loss": 0.4915982186794281, "lr": 3e-05, "grad_norm": 0.2849306084083531, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122220139372, "tokens_per_second": 151828.51635509788, "data_wait_s": 1.1789998097810894e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 835100, "loss": 0.5472451448440552, "lr": 3e-05, "grad_norm": 0.2910525719190749, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122234886213, "tokens_per_second": 139815.18583565488, "data_wait_s": 1.1014002666343004e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 835200, "loss": 0.5077463388442993, "lr": 3e-05, "grad_norm": 0.2840245954475329, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122249570745, "tokens_per_second": 151653.3067558072, "data_wait_s": 9.439994755666703e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f5945fa7ff6d4e74da9ad27cca98f2c603f95925295fef52cea759f51b8204bc
-size 749064
+oid sha256:2469bc2c67aa9a1081e8b75b5097308f738558f372a44f2e48d5a72b52b296c7
+size 751182