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": 76315.86645461465,
-      "required_pages_per_second_upper_estimate": 0.005279708718536894
+      "required_token_positions_per_second": 76617.4622805663,
+      "required_pages_per_second_upper_estimate": 0.0053005738175749585
     },
     {
       "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": 9539.48330682683,
-      "required_pages_per_second_upper_estimate": 0.003993982453504357
+      "required_token_positions_per_second": 9577.182785070787,
+      "required_pages_per_second_upper_estimate": 0.004009766437790474
     },
     {
       "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": 19078.96661365366,
-      "required_pages_per_second_upper_estimate": 0.003973645599439341
+      "required_token_positions_per_second": 19154.365570141574,
+      "required_pages_per_second_upper_estimate": 0.003989349213671576
     },
     {
       "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": 28618.449920480492,
-      "required_pages_per_second_upper_estimate": 0.001715632374889964
+      "required_token_positions_per_second": 28731.54835521236,
+      "required_pages_per_second_upper_estimate": 0.001722412453360829
     },
     {
       "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": 28618.449920480492,
-      "required_pages_per_second_upper_estimate": 0.002224958602059488
+      "required_token_positions_per_second": 28731.54835521236,
+      "required_pages_per_second_upper_estimate": 0.0022337515078924503
     },
     {
       "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": 28618.449920480492,
-      "required_pages_per_second_upper_estimate": 0.3670867474824655
+      "required_token_positions_per_second": 28731.54835521236,
+      "required_pages_per_second_upper_estimate": 0.3685374527675679
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.12175954804672529,
-    "producer_p95_seconds": 0.32890146499994444,
-    "gpu_update_seconds": 0.9660900599726053,
-    "producer_batches_per_second": 8.212908277355378,
-    "gpu_batches_per_second": 1.035100185202564,
-    "producer_headroom": 7.934409050219764,
+    "producer_batch_seconds": 0.12368250403119418,
+    "producer_p95_seconds": 0.34458111799904145,
+    "gpu_update_seconds": 0.962287157593587,
+    "producer_batches_per_second": 8.085217936303975,
+    "gpu_batches_per_second": 1.0391908403939656,
+    "producer_headroom": 7.78030138645064,
     "sustainable": true,
-    "buffer_seconds": 60.86367377827413,
+    "buffer_seconds": 60.62409092839598,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 326592.1575507377,
+    "generation_bytes_per_second": 327141.0607371509,
     "upload_bytes_per_second": 2958687.9373242455,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.07907433406447,
+    "estimated_full_shards_per_hour": 70.19685618005653,
     "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-v1/1791342976705245012-6808dd50/train.jsonl

@@ -413,3 +413,4 @@
 {"step": 883800, "loss": 0.5805310010910034, "lr": 3e-05, "grad_norm": 0.32932567289974446, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129363442301, "tokens_per_second": 152204.26088067185, "data_wait_s": 1.1217998689971864e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 883900, "loss": 0.5239691138267517, "lr": 3e-05, "grad_norm": 0.3012883191103843, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129378229716, "tokens_per_second": 152706.0794318554, "data_wait_s": 1.2616001185961068e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 884000, "loss": 0.515288770198822, "lr": 3e-05, "grad_norm": 0.30551710940741256, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129392852352, "tokens_per_second": 151967.4666830402, "data_wait_s": 1.2092001270502806e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 884100, "loss": 0.6586899757385254, "lr": 3e-05, "grad_norm": 0.607955071258243, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129407429533, "tokens_per_second": 151425.7244361318, "data_wait_s": 1.3840995961800218e-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 sha256:80fdb65292d56123779d1bd642a0760b9b2fdfeff9386691d5fe9a80b8d5ed43
-size 885399
+oid sha256:132748b410944978474d17547ccd91aa3c600e256ea1ced9ec16a1b0c383b6ec
+size 887517