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": 76653.33321129926,
-      "required_pages_per_second_upper_estimate": 0.005303055451795092
+      "required_token_positions_per_second": 76220.92931063878,
+      "required_pages_per_second_upper_estimate": 0.00527314074663212
     },
     {
       "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": 9581.666651412408,
-      "required_pages_per_second_upper_estimate": 0.004011643738993934
+      "required_token_positions_per_second": 9527.616163829847,
+      "required_pages_per_second_upper_estimate": 0.003989013928545645
     },
     {
       "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": 19163.333302824816,
-      "required_pages_per_second_upper_estimate": 0.0039912169558944804
+      "required_token_positions_per_second": 19055.232327659694,
+      "required_pages_per_second_upper_estimate": 0.003968702373582009
     },
     {
       "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": 28744.99995423722,
-      "required_pages_per_second_upper_estimate": 0.0017232188561829669
+      "required_token_positions_per_second": 28582.84849148954,
+      "required_pages_per_second_upper_estimate": 0.001713498123582191
     },
     {
       "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": 28744.99995423722,
-      "required_pages_per_second_upper_estimate": 0.002234797310549302
+      "required_token_positions_per_second": 28582.84849148954,
+      "required_pages_per_second_upper_estimate": 0.0022221907475495784
     },
     {
       "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": 28744.99995423722,
-      "required_pages_per_second_upper_estimate": 0.36870999543665706
+      "required_token_positions_per_second": 28582.84849148954,
+      "required_pages_per_second_upper_estimate": 0.3666300905772058
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.050071186320330696,
-    "producer_p95_seconds": 0.16888301499966474,
-    "gpu_update_seconds": 0.9618368427210411,
-    "producer_batches_per_second": 19.971565954169616,
-    "gpu_batches_per_second": 1.039677371029992,
-    "producer_headroom": 19.20938794155354,
+    "producer_batch_seconds": 0.09931213965618468,
+    "producer_p95_seconds": 0.17139582100026018,
+    "gpu_update_seconds": 0.9672933755441523,
+    "producer_batches_per_second": 10.069262463400413,
+    "gpu_batches_per_second": 1.0338125177766762,
+    "producer_headroom": 9.739930877462612,
     "sustainable": true,
-    "buffer_seconds": 60.59572109142559,
+    "buffer_seconds": 60.939482659281595,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 247234.35305350702,
+    "generation_bytes_per_second": 263599.50626581226,
     "upload_bytes_per_second": 4152236.9850670258,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 53.05073684410007,
+    "estimated_full_shards_per_hour": 56.56231776219154,
     "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-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -55,3 +55,4 @@
 {"step": 725500, "loss": 0.5536834597587585, "lr": 5e-05, "grad_norm": 0.2220945458908904, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106193169137, "tokens_per_second": 152365.59023065472, "data_wait_s": 1.809900004445808e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 725600, "loss": 0.5268424153327942, "lr": 5e-05, "grad_norm": 0.2167088012334988, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106207837305, "tokens_per_second": 151715.4611492302, "data_wait_s": 1.212900042446563e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 725700, "loss": 0.5652655363082886, "lr": 5e-05, "grad_norm": 0.23341700700260484, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106222370869, "tokens_per_second": 150301.46178247128, "data_wait_s": 8.197000170184765e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 725800, "loss": 0.5135230422019958, "lr": 5e-05, "grad_norm": 0.24270620585370845, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106236911957, "tokens_per_second": 150290.77242181444, "data_wait_s": 1.1297999662929215e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:97a7315d71d1cf0cc91efe8f328395c4e77eb0848cc7beaca7ba2d0449349519
-size 121674
+oid sha256:d0130654dd006163f6e705f3e58cf92dc22362b268b93e6c8aa70da92c239a5f
+size 123792