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": 75819.24140740428,
-      "required_pages_per_second_upper_estimate": 0.005245351045442011
+      "required_token_positions_per_second": 77428.20878631044,
+      "required_pages_per_second_upper_estimate": 0.0053566631420334096
     },
     {
       "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": 9477.405175925534,
-      "required_pages_per_second_upper_estimate": 0.003967991636434769
+      "required_token_positions_per_second": 9678.526098288805,
+      "required_pages_per_second_upper_estimate": 0.004052196766745827
     },
     {
       "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": 18954.81035185107,
-      "required_pages_per_second_upper_estimate": 0.003947787124326666
+      "required_token_positions_per_second": 19357.05219657761,
+      "required_pages_per_second_upper_estimate": 0.004031563492752413
     },
     {
       "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": 28432.215527776607,
-      "required_pages_per_second_upper_estimate": 0.0017044679074108187
+      "required_token_positions_per_second": 29035.578294866416,
+      "required_pages_per_second_upper_estimate": 0.0017406385840162512
     },
     {
       "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": 28432.215527776607,
-      "required_pages_per_second_upper_estimate": 0.002210479697185283
+      "required_token_positions_per_second": 29035.578294866416,
+      "required_pages_per_second_upper_estimate": 0.0022573884984141775
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28432.215527776607,
-      "required_pages_per_second_upper_estimate": 0.36469793265577155
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 29035.578294866416,
+      "required_pages_per_second_upper_estimate": 0.37243722239153443
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 62,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.40109595647672336,
-    "producer_p95_seconds": 0.2959250309977506,
-    "gpu_update_seconds": 0.9724180647473472,
-    "producer_batches_per_second": 2.4931689882494057,
-    "gpu_batches_per_second": 1.0283642769016423,
-    "producer_headroom": 2.4244025626415886,
+    "producer_batch_seconds": 0.13620513864844952,
+    "producer_p95_seconds": 0.31357192400173517,
+    "gpu_update_seconds": 0.952211101815329,
+    "producer_batches_per_second": 7.341866906953025,
+    "gpu_batches_per_second": 1.050187293651129,
+    "producer_headroom": 6.991007176851241,
     "sustainable": true,
-    "buffer_seconds": 60.28992001433552,
+    "buffer_seconds": 60.941510516181054,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 306197.5668873526,
-    "upload_bytes_per_second": 5609868.625905031,
+    "generation_bytes_per_second": 334942.6797940333,
+    "upload_bytes_per_second": 5615689.179757893,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.70286993947443,
+    "estimated_full_shards_per_hour": 71.87090201726674,
     "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": 1.5769093039998552,
-      "bytes": 5962126,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.0608517170003324,
       "bytes": 4607080,
@@ -562,6 +556,12 @@
       "bytes": 6042585,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.5961815709997609,
+      "bytes": 6520796,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -306,3 +306,4 @@
 {"step": 675600, "loss": 0.5709023475646973, "lr": 0.0001, "grad_norm": 0.17599447618569133, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98890321929, "tokens_per_second": 153550.92980253155, "data_wait_s": 1.4754998119315132e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 675700, "loss": 0.5651541352272034, "lr": 0.0001, "grad_norm": 0.19614258731955755, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98905083073, "tokens_per_second": 153414.86164814638, "data_wait_s": 8.637998689664528e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 675800, "loss": 0.5303199291229248, "lr": 0.0001, "grad_norm": 0.19935096478378747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98919667570, "tokens_per_second": 152462.89387819162, "data_wait_s": 1.1382002412574366e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 675900, "loss": 0.5435974597930908, "lr": 0.0001, "grad_norm": 0.19091877172449642, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98934190928, "tokens_per_second": 151262.43462264814, "data_wait_s": 1.371300095343031e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bb5204e011b3261c42d471286cdb9d352ffa65eb462944c65bed5c0d12a3d2fb
-size 657207
+oid sha256:748ce251e60f28f241e04c06aa9a8331d7b8e93acf4520d17498ef5a4818bd07
+size 659325