Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75117.40374702057,
-      "required_pages_per_second_upper_estimate": 0.006510986173472515
+      "required_token_positions_per_second": 74814.24155671152,
+      "required_pages_per_second_upper_estimate": 0.006484708843174074
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9389.67546837757,
-      "required_pages_per_second_upper_estimate": 0.003745681932494643
+      "required_token_positions_per_second": 9351.78019458894,
+      "required_pages_per_second_upper_estimate": 0.0037305649411955243
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18779.35093675514,
-      "required_pages_per_second_upper_estimate": 0.004344272333929972
+      "required_token_positions_per_second": 18703.56038917788,
+      "required_pages_per_second_upper_estimate": 0.004326739524616053
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28169.026405132714,
-      "required_pages_per_second_upper_estimate": 0.0017153459166129603
+      "required_token_positions_per_second": 28055.340583766818,
+      "required_pages_per_second_upper_estimate": 0.001708423046555185
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28169.026405132714,
-      "required_pages_per_second_upper_estimate": 0.0021685296295229466
+      "required_token_positions_per_second": 28055.340583766818,
+      "required_pages_per_second_upper_estimate": 0.0021597777802917044
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28169.026405132714,
-      "required_pages_per_second_upper_estimate": 0.3497650322849462
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28055.340583766818,
+      "required_pages_per_second_upper_estimate": 0.3483534348655502
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.1084905962889593,
-    "producer_p95_seconds": 0.32796435699856374,
-    "gpu_update_seconds": 0.9815035707077978,
-    "producer_batches_per_second": 9.217388734195449,
-    "gpu_batches_per_second": 1.018844994398608,
-    "producer_headroom": 9.046899955214663,
+    "producer_batch_seconds": 0.5097604426248381,
+    "producer_p95_seconds": 0.343780216997402,
+    "gpu_update_seconds": 0.9854808184363119,
+    "producer_batches_per_second": 1.9617057668320437,
+    "gpu_batches_per_second": 1.0147330940309178,
+    "producer_headroom": 1.933223404628875,
     "sustainable": true,
-    "buffer_seconds": 60.853221383883465,
+    "buffer_seconds": 60.11432992461502,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 405661.6745687513,
+    "generation_bytes_per_second": 364593.27780105924,
     "upload_bytes_per_second": 5523087.894636441,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.04555204197793,
+    "estimated_full_shards_per_hour": 78.23323011897881,
     "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-v1/1790563279934888633-82973089/train.jsonl

@@ -227,3 +227,4 @@
 {"step": 177700, "loss": 0.5706491265445948, "lr": 0.0003, "grad_norm": 0.15813942364198907, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26007069522, "tokens_per_second": 147803.53629690854, "data_wait_s": 1.3288001355249435e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 177800, "loss": 0.5302779879420996, "lr": 0.0003, "grad_norm": 0.14627401276772195, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26021858707, "tokens_per_second": 151587.71856583096, "data_wait_s": 1.1870000889757648e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 177900, "loss": 0.5778142791241407, "lr": 0.0003, "grad_norm": 0.15374156595544317, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26036536685, "tokens_per_second": 149747.60026334377, "data_wait_s": 1.2282001989660785e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 178000, "loss": 0.6353771220892668, "lr": 0.0003, "grad_norm": 0.18181807817259837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26051312650, "tokens_per_second": 151056.3295164474, "data_wait_s": 1.0265001037623733e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ea2d19c6a06fcddf6a4e3ef3ad4218999dd67b940ae0053d26412eade0b7336f
-size 488319
+oid sha256:828c41f02e615ec3cbf81c599614d454542cf2d1ed3286a41161e44f73f0b8fb
+size 490437