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": 77608.3541749895,
-      "required_pages_per_second_upper_estimate": 0.0053691260180171166
+      "required_token_positions_per_second": 77801.65125515989,
+      "required_pages_per_second_upper_estimate": 0.005382498758534323
     },
     {
       "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": 9701.044271873687,
-      "required_pages_per_second_upper_estimate": 0.004061624655792881
+      "required_token_positions_per_second": 9725.206406894986,
+      "required_pages_per_second_upper_estimate": 0.004071740837163543
     },
     {
       "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": 19402.088543747373,
-      "required_pages_per_second_upper_estimate": 0.004040943376179531
+      "required_token_positions_per_second": 19450.41281378997,
+      "required_pages_per_second_upper_estimate": 0.004051008047232704
     },
     {
       "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": 29103.132815621062,
-      "required_pages_per_second_upper_estimate": 0.0017446883743857118
+      "required_token_positions_per_second": 29175.61922068496,
+      "required_pages_per_second_upper_estimate": 0.0017490338236889605
     },
     {
       "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": 29103.132815621062,
-      "required_pages_per_second_upper_estimate": 0.0022626405652619174
+      "required_token_positions_per_second": 29175.61922068496,
+      "required_pages_per_second_upper_estimate": 0.0022682760644216348
     },
     {
       "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": 29103.132815621062,
-      "required_pages_per_second_upper_estimate": 0.37330373924938187
+      "required_token_positions_per_second": 29175.61922068496,
+      "required_pages_per_second_upper_estimate": 0.3742335170237036
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.0943975838359421,
-    "producer_p95_seconds": 0.1857063029983692,
-    "gpu_update_seconds": 0.9500008186458875,
-    "producer_batches_per_second": 10.593491478955075,
-    "gpu_batches_per_second": 1.052630671861294,
-    "producer_headroom": 10.063825577325554,
+    "producer_batch_seconds": 0.0506923616640691,
+    "producer_p95_seconds": 0.17497649699907925,
+    "gpu_update_seconds": 0.9476405553167521,
+    "producer_batches_per_second": 19.726837874053974,
+    "gpu_batches_per_second": 1.0552524313037095,
+    "producer_headroom": 18.693951597612045,
     "sustainable": true,
-    "buffer_seconds": 60.8000523933368,
+    "buffer_seconds": 60.64899554027213,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 329508.9777968561,
-    "upload_bytes_per_second": 4154890.614714338,
+    "generation_bytes_per_second": 355412.8582836409,
+    "upload_bytes_per_second": 4178714.2890595957,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.70495605878126,
+    "estimated_full_shards_per_hour": 76.26332579976959,
     "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": [
@@ -436,6 +436,12 @@
       "bytes": 1276481,
       "pages": 36,
       "shards": 1
+    },
+    {
+      "seconds": 1.4530193070004316,
+      "bytes": 7690650,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -69,3 +69,4 @@
 {"step": 531900, "loss": 0.639557957649231, "lr": 0.0003, "grad_norm": 0.180102644276118, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77853197157, "tokens_per_second": 153762.32467936815, "data_wait_s": 1.095200059353374e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 532000, "loss": 0.624271035194397, "lr": 0.0003, "grad_norm": 0.1854685226702973, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77867764813, "tokens_per_second": 152493.90096277112, "data_wait_s": 1.7542999557917938e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 532100, "loss": 0.6729798913002014, "lr": 0.0003, "grad_norm": 0.16285817975727232, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77882264539, "tokens_per_second": 152094.4709972879, "data_wait_s": 1.054499989550095e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 532200, "loss": 0.6312143802642822, "lr": 0.0003, "grad_norm": 0.17787364407634543, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77896848290, "tokens_per_second": 143940.44404040152, "data_wait_s": 1.2266998965060338e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eae9660a168d3d4061232deac48d6dbafe63a6231ccfa5cd62a94b5ca4e24e6e
-size 151326
+oid sha256:c4b74362a7173536d9a6258b7d453508d5e3600632579457b3f09f381d3f9dd1
+size 153444