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": 77145.50735235338,
-      "required_pages_per_second_upper_estimate": 0.00533710520087456
+      "required_token_positions_per_second": 76882.51112516246,
+      "required_pages_per_second_upper_estimate": 0.005318910511642171
     },
     {
       "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": 9643.188419044172,
-      "required_pages_per_second_upper_estimate": 0.004037401618380755
+      "required_token_positions_per_second": 9610.313890645308,
+      "required_pages_per_second_upper_estimate": 0.004023637739838368
     },
     {
       "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": 19286.376838088345,
-      "required_pages_per_second_upper_estimate": 0.00401684367941364
+      "required_token_positions_per_second": 19220.627781290616,
+      "required_pages_per_second_upper_estimate": 0.004003149884801902
     },
     {
       "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": 28929.56525713252,
-      "required_pages_per_second_upper_estimate": 0.0017342832642766402
+      "required_token_positions_per_second": 28830.94167193592,
+      "required_pages_per_second_upper_estimate": 0.0017283709309335992
     },
     {
       "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": 28929.56525713252,
-      "required_pages_per_second_upper_estimate": 0.0022491464510324397
+      "required_token_positions_per_second": 28830.94167193592,
+      "required_pages_per_second_upper_estimate": 0.00224147890108271
     },
     {
       "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": 28929.56525713252,
-      "required_pages_per_second_upper_estimate": 0.37107740097141545
+      "required_token_positions_per_second": 28830.94167193592,
+      "required_pages_per_second_upper_estimate": 0.36981236351426894
     }
   ],
   "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": 0.09404523918752261,
-    "producer_p95_seconds": 0.275779796000279,
-    "gpu_update_seconds": 0.9557005006559318,
-    "producer_batches_per_second": 10.633180463351668,
-    "gpu_batches_per_second": 1.0463529100525362,
-    "producer_headroom": 10.162135892390062,
+    "producer_batch_seconds": 0.11320519675800256,
+    "producer_p95_seconds": 0.29611602099976153,
+    "gpu_update_seconds": 0.9589697178331362,
+    "producer_batches_per_second": 8.833516734551404,
+    "gpu_batches_per_second": 1.0427857954259232,
+    "producer_headroom": 8.471075050407046,
     "sustainable": true,
-    "buffer_seconds": 60.20913154132371,
+    "buffer_seconds": 60.415092223487584,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 362631.84989122016,
+    "generation_bytes_per_second": 406461.5631673876,
     "upload_bytes_per_second": 4837492.416788516,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.81235334923223,
+    "estimated_full_shards_per_hour": 87.2171895147917,
     "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-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -268,3 +268,4 @@
 {"step": 786800, "loss": 0.48189279437065125, "lr": 3e-05, "grad_norm": 0.24508727685326637, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115165046750, "tokens_per_second": 151457.01707543465, "data_wait_s": 1.237800097442232e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 786900, "loss": 0.5313212871551514, "lr": 3e-05, "grad_norm": 0.25772295429552444, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115179794041, "tokens_per_second": 153177.10567686593, "data_wait_s": 1.2923999747727066e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 787000, "loss": 0.48010745644569397, "lr": 3e-05, "grad_norm": 0.24818863639062674, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115194548028, "tokens_per_second": 153319.04305843593, "data_wait_s": 1.7002999811666086e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 787100, "loss": 0.5606993436813354, "lr": 3e-05, "grad_norm": 0.2726448551459696, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115209024455, "tokens_per_second": 149759.58569847242, "data_wait_s": 1.3065000530332327e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eaf5fd75aa32fe9c39401e31f4c806799882233309f15293ee3bc3d1785008bc
-size 575940
+oid sha256:2a7694f7ae67e6eaa8295feae1381d3affafc61dac137764717c320104d8be71
+size 578058