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": 76039.94462746628,
-      "required_pages_per_second_upper_estimate": 0.005260619806307925
+      "required_token_positions_per_second": 77232.34058752313,
+      "required_pages_per_second_upper_estimate": 0.0053431125255644675
     },
     {
       "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": 9504.993078433285,
-      "required_pages_per_second_upper_estimate": 0.003979542115113849
+      "required_token_positions_per_second": 9654.04257344039,
+      "required_pages_per_second_upper_estimate": 0.004041946026165933
     },
     {
       "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": 19009.98615686657,
-      "required_pages_per_second_upper_estimate": 0.003959278789427566
+      "required_token_positions_per_second": 19308.08514688078,
+      "required_pages_per_second_upper_estimate": 0.0040213649476485034
     },
     {
       "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": 28514.979235299852,
-      "required_pages_per_second_upper_estimate": 0.001709429465303973
+      "required_token_positions_per_second": 28962.127720321172,
+      "required_pages_per_second_upper_estimate": 0.0017362353342247974
     },
     {
       "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": 28514.979235299852,
-      "required_pages_per_second_upper_estimate": 0.002216914211406139
+      "required_token_positions_per_second": 28962.127720321172,
+      "required_pages_per_second_upper_estimate": 0.002251678039318219
     },
     {
       "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": 28514.979235299852,
-      "required_pages_per_second_upper_estimate": 0.3657595366311342
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28962.127720321172,
+      "required_pages_per_second_upper_estimate": 0.3714950772863505
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05139413005469606,
-    "producer_p95_seconds": 0.15724994399988645,
-    "gpu_update_seconds": 0.969595656088482,
-    "producer_batches_per_second": 19.457474986652226,
-    "gpu_batches_per_second": 1.0313577559063893,
-    "producer_headroom": 18.86588322550829,
+    "producer_batch_seconds": 0.09985365482813435,
+    "producer_p95_seconds": 0.16238798999984283,
+    "gpu_update_seconds": 0.9546259952648741,
+    "producer_batches_per_second": 10.014655965484442,
+    "gpu_batches_per_second": 1.047530661180598,
+    "producer_headroom": 9.560250918285893,
     "sustainable": true,
-    "buffer_seconds": 60.11493067748588,
+    "buffer_seconds": 60.14143770168707,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 403027.47604566225,
-    "upload_bytes_per_second": 3951706.6780117266,
+    "generation_bytes_per_second": 401650.28783090686,
+    "upload_bytes_per_second": 3811035.8869786924,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.48031435992623,
+    "estimated_full_shards_per_hour": 86.18480182834057,
     "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": [
@@ -304,6 +304,18 @@
       "bytes": 10661870,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.9814531229999375,
+      "bytes": 6026004,
+      "pages": 34,
+      "shards": 2
+    },
+    {
+      "seconds": 1.2945359899999858,
+      "bytes": 2043786,
+      "pages": 19,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -23,3 +23,4 @@
 {"step": 844800, "loss": 0.4664148986339569, "lr": 3e-05, "grad_norm": 0.29752736495565235, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123654938761, "tokens_per_second": 150062.9922451329, "data_wait_s": 1.7485999705968425e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 844900, "loss": 0.48894795775413513, "lr": 3e-05, "grad_norm": 0.29625907752383457, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123669544844, "tokens_per_second": 151249.47469693146, "data_wait_s": 1.471799987484701e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 845000, "loss": 0.5380342602729797, "lr": 3e-05, "grad_norm": 0.32369832958366457, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123684316210, "tokens_per_second": 151915.52533154108, "data_wait_s": 1.2603999948623823e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 845100, "loss": 0.5290372371673584, "lr": 3e-05, "grad_norm": 0.30341436897954355, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123699004752, "tokens_per_second": 138643.06287528094, "data_wait_s": 1.079499998013489e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:312e6be3219d175badf556b7c5f31b8b141c81e92e67fb26b60b891c13f2b028
-size 53898
+oid sha256:fcb0d4987cf75d69c4d0f8277060d98ba6378a187d60f688e507ad62dda5b23b
+size 56016