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": 77157.8593802369,
-      "required_pages_per_second_upper_estimate": 0.005337959742824195
+      "required_token_positions_per_second": 76676.64163272269,
+      "required_pages_per_second_upper_estimate": 0.005304667982472154
     },
     {
       "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": 9644.732422529612,
-      "required_pages_per_second_upper_estimate": 0.00403804806039765
+      "required_token_positions_per_second": 9584.580204090336,
+      "required_pages_per_second_upper_estimate": 0.004012863582658285
     },
     {
       "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": 19289.464845059225,
-      "required_pages_per_second_upper_estimate": 0.004017486829829454
+      "required_token_positions_per_second": 19169.160408180673,
+      "required_pages_per_second_upper_estimate": 0.003992430588269003
     },
     {
       "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": 28934.197267588836,
-      "required_pages_per_second_upper_estimate": 0.001734560946230826
+      "required_token_positions_per_second": 28753.74061227101,
+      "required_pages_per_second_upper_estimate": 0.0017237428452858805
     },
     {
       "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": 28934.197267588836,
-      "required_pages_per_second_upper_estimate": 0.0022495065694713577
+      "required_token_positions_per_second": 28753.74061227101,
+      "required_pages_per_second_upper_estimate": 0.002235476858265338
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28934.197267588836,
-      "required_pages_per_second_upper_estimate": 0.37113681542808374
+      "required_token_positions_per_second": 28753.74061227101,
+      "required_pages_per_second_upper_estimate": 0.3688221112129271
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.09026600030443888,
-    "producer_p95_seconds": 0.3296432319984888,
-    "gpu_update_seconds": 0.9555475047158266,
-    "producer_batches_per_second": 11.078368340541443,
-    "gpu_batches_per_second": 1.0465204451529526,
-    "producer_headroom": 10.58590722412719,
+    "producer_batch_seconds": 0.10704750380483574,
+    "producer_p95_seconds": 0.32344004300102824,
+    "gpu_update_seconds": 0.9615444603475654,
+    "producer_batches_per_second": 9.34164706748469,
+    "gpu_batches_per_second": 1.0399935117285521,
+    "producer_headroom": 8.982408988261982,
     "sustainable": true,
-    "buffer_seconds": 60.19949279709708,
+    "buffer_seconds": 60.577301001896615,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 474060.19653001765,
+    "generation_bytes_per_second": 493875.72651441075,
     "upload_bytes_per_second": 5921794.24519696,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.72228261876485,
+    "estimated_full_shards_per_hour": 105.97423407148592,
     "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-v1/1790914434605211575-090606d4/train.jsonl

@@ -323,3 +323,4 @@
 {"step": 517300, "loss": 0.5869436860084534, "lr": 0.0003, "grad_norm": 0.158736958374673, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75714707979, "tokens_per_second": 150045.42190867095, "data_wait_s": 2.1870997443329543e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 517400, "loss": 0.6182101368904114, "lr": 0.0003, "grad_norm": 0.17699378061598395, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75729315612, "tokens_per_second": 152490.06762253493, "data_wait_s": 1.4882003597449511e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 517500, "loss": 0.6253323554992676, "lr": 0.0003, "grad_norm": 0.16925463581210798, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75744017905, "tokens_per_second": 153063.1224845746, "data_wait_s": 1.756600249791518e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 517600, "loss": 0.5789720416069031, "lr": 0.0003, "grad_norm": 0.18343476660004074, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75758639250, "tokens_per_second": 152493.2833605163, "data_wait_s": 9.866998880170286e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fa2e66a97f2906f5bb6303cbb7b8feb5387bb3f21ee96d3641ccc92c7238e381
-size 693213
+oid sha256:44a74cc2c44093377e055453c88ca8e62195be4424833ebe7e41998409d2b0e2
+size 695331