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": 76423.15774627705,
-      "required_pages_per_second_upper_estimate": 0.005287131379044194
+      "required_token_positions_per_second": 76664.66716803999,
+      "required_pages_per_second_upper_estimate": 0.005303839561220818
     },
     {
       "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": 9552.894718284631,
-      "required_pages_per_second_upper_estimate": 0.00399959753141962
+      "required_token_positions_per_second": 9583.083396004999,
+      "required_pages_per_second_upper_estimate": 0.004012236900369861
     },
     {
       "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": 19105.789436569263,
-      "required_pages_per_second_upper_estimate": 0.003979232086087251
+      "required_token_positions_per_second": 19166.166792009997,
+      "required_pages_per_second_upper_estimate": 0.003991807096967619
     },
     {
       "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": 28658.684154853894,
-      "required_pages_per_second_upper_estimate": 0.0017180443558065309
+      "required_token_positions_per_second": 28749.250188014994,
+      "required_pages_per_second_upper_estimate": 0.0017234736512082131
     },
     {
       "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": 28658.684154853894,
-      "required_pages_per_second_upper_estimate": 0.002228086636810326
+      "required_token_positions_per_second": 28749.250188014994,
+      "required_pages_per_second_upper_estimate": 0.0022351277475307213
     },
     {
       "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": 28658.684154853894,
-      "required_pages_per_second_upper_estimate": 0.3676028290408524
+      "required_token_positions_per_second": 28749.250188014994,
+      "required_pages_per_second_upper_estimate": 0.36876451287201284
     }
   ],
   "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": 0.10814964682032979,
-    "producer_p95_seconds": 0.20681603899993206,
-    "gpu_update_seconds": 0.9647337557651713,
-    "producer_batches_per_second": 9.246447209035376,
-    "gpu_batches_per_second": 1.0365554164805373,
-    "producer_headroom": 8.920359743457084,
+    "producer_batch_seconds": 0.053477873406244214,
+    "producer_p95_seconds": 0.18304753299980803,
+    "gpu_update_seconds": 0.9616946466146764,
+    "producer_batches_per_second": 18.699322473119086,
+    "gpu_batches_per_second": 1.0398310976567924,
+    "producer_headroom": 17.983038317720137,
     "sustainable": true,
-    "buffer_seconds": 60.77822661320579,
+    "buffer_seconds": 60.586762736724616,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 279925.1247667336,
-    "upload_bytes_per_second": 5268727.554680633,
+    "generation_bytes_per_second": 298493.2691214525,
+    "upload_bytes_per_second": 5256151.386890771,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.06541545154101,
+    "estimated_full_shards_per_hour": 64.0497069857853,
     "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": [
@@ -496,6 +496,12 @@
       "bytes": 2479743,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.3296366939994186,
+      "bytes": 6147791,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -77,3 +77,4 @@
 {"step": 892700, "loss": 0.5476626753807068, "lr": 3e-05, "grad_norm": 0.3149015245871685, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130666031796, "tokens_per_second": 153026.12313189296, "data_wait_s": 1.1962999451498035e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 892800, "loss": 0.5209189057350159, "lr": 3e-05, "grad_norm": 0.31463446748906326, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130680691402, "tokens_per_second": 153371.75116730493, "data_wait_s": 1.1312999959045555e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 892900, "loss": 0.5386309027671814, "lr": 3e-05, "grad_norm": 0.3150166992141552, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130695448725, "tokens_per_second": 154076.71778541556, "data_wait_s": 1.0897000720433425e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 893000, "loss": 0.5396263003349304, "lr": 3e-05, "grad_norm": 0.30559744526303734, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130710170691, "tokens_per_second": 154258.12412871476, "data_wait_s": 1.2008000339847058e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:038da0579c5f787684e25a779af6208254356dedeb652c1834749932ac63f610
-size 168270
+oid sha256:3d935617aa2cfaea1d239cf263444e15413fa8ebc30fc64ac841b50dfbfe5981
+size 170388