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": 76478.38649550175,
-      "required_pages_per_second_upper_estimate": 0.005290952232063913
+      "required_token_positions_per_second": 76102.9871783626,
+      "required_pages_per_second_upper_estimate": 0.005264981236257803
     },
     {
       "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": 9559.798311937719,
-      "required_pages_per_second_upper_estimate": 0.004002487921918739
+      "required_token_positions_per_second": 9512.873397295325,
+      "required_pages_per_second_upper_estimate": 0.003982841440061615
     },
     {
       "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": 19119.596623875437,
-      "required_pages_per_second_upper_estimate": 0.003982107759083113
+      "required_token_positions_per_second": 19025.74679459065,
+      "required_pages_per_second_upper_estimate": 0.0039625613146295246
     },
     {
       "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": 28679.394935813154,
-      "required_pages_per_second_upper_estimate": 0.001719285935501507
+      "required_token_positions_per_second": 28538.620191885973,
+      "required_pages_per_second_upper_estimate": 0.0017108467045536604
     },
     {
       "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": 28679.394935813154,
-      "required_pages_per_second_upper_estimate": 0.0022296968089328107
+      "required_token_positions_per_second": 28538.620191885973,
+      "required_pages_per_second_upper_estimate": 0.0022187521917951315
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28679.394935813154,
-      "required_pages_per_second_upper_estimate": 0.36786848470149375
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28538.620191885973,
+      "required_pages_per_second_upper_estimate": 0.36606277743853943
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0753768615235515,
-    "producer_p95_seconds": 0.21246311299910303,
-    "gpu_update_seconds": 0.9640370747666922,
-    "producer_batches_per_second": 13.266670696916057,
-    "gpu_batches_per_second": 1.0373045043335198,
-    "producer_headroom": 12.789562410547948,
+    "producer_batch_seconds": 0.11645054338279692,
+    "producer_p95_seconds": 0.21987846000047284,
+    "gpu_update_seconds": 0.9687924578729565,
+    "producer_batches_per_second": 8.587336485951758,
+    "gpu_batches_per_second": 1.0322128252273572,
+    "producer_headroom": 8.31934682080732,
     "sustainable": true,
-    "buffer_seconds": 60.73433571030161,
+    "buffer_seconds": 60.0651323881233,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 413277.0721094184,
+    "generation_bytes_per_second": 455562.82642948866,
     "upload_bytes_per_second": 5209954.494958308,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.67963907682336,
+    "estimated_full_shards_per_hour": 97.75317759192939,
     "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-v1/1791039222312944374-63c10d34/train.jsonl

@@ -182,3 +182,4 @@
 {"step": 622600, "loss": 0.5647789835929871, "lr": 0.0001, "grad_norm": 0.16720628825279638, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91131949369, "tokens_per_second": 152741.46973627643, "data_wait_s": 9.515999408904463e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 622700, "loss": 0.5370277762413025, "lr": 0.0001, "grad_norm": 0.14730485395166398, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91146648237, "tokens_per_second": 153021.0069713086, "data_wait_s": 1.3280998246045783e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 622800, "loss": 0.5662478804588318, "lr": 0.0001, "grad_norm": 0.16289424166216349, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91161265801, "tokens_per_second": 152186.00582515876, "data_wait_s": 1.0527000995352864e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 622900, "loss": 0.512753427028656, "lr": 0.0001, "grad_norm": 0.16105744409163986, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91175779382, "tokens_per_second": 150315.8605156014, "data_wait_s": 1.1536998499650508e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9737c48ca8e215ad967d699a3f9e04b305f010c16e5bfb74eb851f400d444780
-size 393009
+oid sha256:106acc5b90f6bec1b866232add722e3af8f1055842297ead5e155758be82cce1
+size 395127