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": 76702.70264228176,
-      "required_pages_per_second_upper_estimate": 0.005306470943583323
+      "required_token_positions_per_second": 76977.79696229403,
+      "required_pages_per_second_upper_estimate": 0.005325502606948536
     },
     {
       "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": 9587.83783028522,
-      "required_pages_per_second_upper_estimate": 0.004014227482719112
+      "required_token_positions_per_second": 9622.224620286754,
+      "required_pages_per_second_upper_estimate": 0.004028624513614923
     },
     {
       "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": 19175.67566057044,
-      "required_pages_per_second_upper_estimate": 0.003993787543523032
+      "required_token_positions_per_second": 19244.449240573507,
+      "required_pages_per_second_upper_estimate": 0.004008111266556401
     },
     {
       "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": 28763.51349085566,
-      "required_pages_per_second_upper_estimate": 0.0017243287144347087
+      "required_token_positions_per_second": 28866.673860860257,
+      "required_pages_per_second_upper_estimate": 0.0017305130210997198
     },
     {
       "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": 28763.51349085566,
-      "required_pages_per_second_upper_estimate": 0.0022362366565889446
+      "required_token_positions_per_second": 28866.673860860257,
+      "required_pages_per_second_upper_estimate": 0.0022442569216022884
     },
     {
       "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": 28763.51349085566,
-      "required_pages_per_second_upper_estimate": 0.36894746720611155
+      "required_token_positions_per_second": 28866.673860860257,
+      "required_pages_per_second_upper_estimate": 0.37027069766755505
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.3635166038357909,
-    "producer_p95_seconds": 0.24673293700107024,
-    "gpu_update_seconds": 0.9612177597423799,
-    "producer_batches_per_second": 2.750905981867403,
-    "gpu_batches_per_second": 1.0403469867931012,
-    "producer_headroom": 2.644219685152497,
+    "producer_batch_seconds": 0.11559402712498468,
+    "producer_p95_seconds": 0.23333685100078583,
+    "gpu_update_seconds": 0.9577826712306944,
+    "producer_batches_per_second": 8.650966013311066,
+    "gpu_batches_per_second": 1.0440781923054203,
+    "producer_headroom": 8.285745336955024,
     "sustainable": true,
-    "buffer_seconds": 60.55671886376993,
+    "buffer_seconds": 60.340308287533745,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 323291.7802778824,
-    "upload_bytes_per_second": 5376747.698464667,
+    "generation_bytes_per_second": 344454.5656583641,
+    "upload_bytes_per_second": 5447442.886427647,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.37089020016055,
+    "estimated_full_shards_per_hour": 73.91193129838173,
     "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": 1.0752750450001258,
-      "bytes": 6242403,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 0.8482705019996502,
       "bytes": 2583323,
@@ -562,6 +556,12 @@
       "bytes": 2661454,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.082202919998963,
+      "bytes": 11730331,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -134,3 +134,4 @@
 {"step": 410500, "loss": 0.5681799054145813, "lr": 0.0003, "grad_norm": 0.1623207670027057, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60086465650, "tokens_per_second": 150977.93994714442, "data_wait_s": 2.4000999474083073e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 410600, "loss": 0.6016080975532532, "lr": 0.0003, "grad_norm": 0.1727383479707727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60101196151, "tokens_per_second": 152161.6338070798, "data_wait_s": 1.414700091117993e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 410700, "loss": 0.6070379018783569, "lr": 0.0003, "grad_norm": 0.16094560335790287, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60115921087, "tokens_per_second": 152569.10342319703, "data_wait_s": 1.0637000741553493e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 410800, "loss": 0.618433952331543, "lr": 0.0003, "grad_norm": 0.15794403351012848, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60130655350, "tokens_per_second": 152182.62973155212, "data_wait_s": 1.8478998754289933e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:738bb476ec6833e0c68e20d8f6d6660602f7fa342c4538658bd5edd44ce6e513
-size 290562
+oid sha256:3f67a06b0ab69ca19fda84c72d2a1fe7871f501852ee4b17dadc650ee5cb7a44
+size 292680