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": 76372.44038302057,
-      "required_pages_per_second_upper_estimate": 0.0052836226341736155
+      "required_token_positions_per_second": 76029.03751020682,
+      "required_pages_per_second_upper_estimate": 0.005259865226627917
     },
     {
       "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": 9546.555047877571,
-      "required_pages_per_second_upper_estimate": 0.003996943243807556
+      "required_token_positions_per_second": 9503.629688775853,
+      "required_pages_per_second_upper_estimate": 0.003978971292335096
     },
     {
       "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": 19093.110095755143,
-      "required_pages_per_second_upper_estimate": 0.003976591313772373
+      "required_token_positions_per_second": 19007.259377551705,
+      "required_pages_per_second_upper_estimate": 0.003958710873206286
     },
     {
       "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": 28639.665143632716,
-      "required_pages_per_second_upper_estimate": 0.0017169041951241699
+      "required_token_positions_per_second": 28510.88906632756,
+      "required_pages_per_second_upper_estimate": 0.0017091842659194105
     },
     {
       "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": 28639.665143632716,
-      "required_pages_per_second_upper_estimate": 0.002226607992343669
+      "required_token_positions_per_second": 28510.88906632756,
+      "required_pages_per_second_upper_estimate": 0.0022165962187592953
     },
     {
       "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": 28639.665143632716,
-      "required_pages_per_second_upper_estimate": 0.36735887358593033
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28510.88906632756,
+      "required_pages_per_second_upper_estimate": 0.3657070723352389
     }
   ],
   "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": 0.13385324279670385,
-    "producer_p95_seconds": 0.2571261470002355,
-    "gpu_update_seconds": 0.9653744155645902,
-    "producer_batches_per_second": 7.470868685032897,
-    "gpu_batches_per_second": 1.0358675182158823,
-    "producer_headroom": 7.2121854905734315,
+    "producer_batch_seconds": 0.0979484652266649,
+    "producer_p95_seconds": 0.25357651299782447,
+    "gpu_update_seconds": 0.9697347541733918,
+    "producer_batches_per_second": 10.20945042564859,
+    "gpu_batches_per_second": 1.0312098186605743,
+    "producer_headroom": 9.900458898761766,
     "sustainable": true,
-    "buffer_seconds": 60.818588180569186,
+    "buffer_seconds": 60.123554758750295,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 288265.61806730065,
+    "generation_bytes_per_second": 276669.0431801136,
     "upload_bytes_per_second": 4994296.893667651,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.85509115709557,
+    "estimated_full_shards_per_hour": 59.36673614075237,
     "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-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -182,3 +182,4 @@
 {"step": 738200, "loss": 0.5587160587310791, "lr": 5e-05, "grad_norm": 0.2236781793592614, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108052271803, "tokens_per_second": 132744.06884565984, "data_wait_s": 1.8234000890515745e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 738300, "loss": 0.5340388417243958, "lr": 5e-05, "grad_norm": 0.2364735244449206, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108067046735, "tokens_per_second": 152113.6151064188, "data_wait_s": 2.4574001145083457e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 738400, "loss": 0.5231951475143433, "lr": 5e-05, "grad_norm": 0.24424534864142355, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108081583373, "tokens_per_second": 148619.50765769536, "data_wait_s": 1.7316000594291836e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 738500, "loss": 0.5642023086547852, "lr": 5e-05, "grad_norm": 0.23152752564406837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108096143363, "tokens_per_second": 149230.83097907345, "data_wait_s": 1.7384001694153994e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cd27927cf22306f691f829b69439f83262c46896db4f3595417cb8b95c16a239
-size 392226
+oid sha256:f3fb7c024254a5db8117d47d2fcbf4bb0905ea2504ca4efee1671597f4e2d089
+size 394344