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": 77247.35536445587,
-      "required_pages_per_second_upper_estimate": 0.005344151282671757
+      "required_token_positions_per_second": 76719.61842918168,
+      "required_pages_per_second_upper_estimate": 0.005307641216971092
     },
     {
       "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": 9655.919420556984,
-      "required_pages_per_second_upper_estimate": 0.004042731822860627
+      "required_token_positions_per_second": 9589.95230364771,
+      "required_pages_per_second_upper_estimate": 0.004015112768560761
     },
     {
       "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": 19311.838841113968,
-      "required_pages_per_second_upper_estimate": 0.0040221467431657025
+      "required_token_positions_per_second": 19179.90460729542,
+      "required_pages_per_second_upper_estimate": 0.00399466832160102
     },
     {
       "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": 28967.758261670948,
-      "required_pages_per_second_upper_estimate": 0.0017365728765813776
+      "required_token_positions_per_second": 28769.85691094313,
+      "required_pages_per_second_upper_estimate": 0.0017247089927831109
     },
     {
       "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": 28967.758261670948,
-      "required_pages_per_second_upper_estimate": 0.0022521157891420298
+      "required_token_positions_per_second": 28769.85691094313,
+      "required_pages_per_second_upper_estimate": 0.002236729829598989
     },
     {
       "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": 28967.758261670948,
-      "required_pages_per_second_upper_estimate": 0.37156729982518116
+      "required_token_positions_per_second": 28769.85691094313,
+      "required_pages_per_second_upper_estimate": 0.36902883378795975
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10741447028900097,
-    "producer_p95_seconds": 0.3103208590000577,
-    "gpu_update_seconds": 0.9544404420338869,
-    "producer_batches_per_second": 9.309732639461687,
-    "gpu_batches_per_second": 1.0477343121264087,
-    "producer_headroom": 8.885585335625118,
+    "producer_batch_seconds": 0.5491582239063462,
+    "producer_p95_seconds": 0.3228149940005096,
+    "gpu_update_seconds": 0.961005822364156,
+    "producer_batches_per_second": 1.8209688145734493,
+    "gpu_batches_per_second": 1.0405764218367741,
+    "producer_headroom": 1.7499616331486398,
     "sustainable": true,
-    "buffer_seconds": 60.129747848134876,
+    "buffer_seconds": 60.543366808941826,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 374144.1236686113,
-    "upload_bytes_per_second": 4958616.861179847,
+    "generation_bytes_per_second": 338107.8012451105,
+    "upload_bytes_per_second": 5034847.288480273,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.28261930984263,
+    "estimated_full_shards_per_hour": 72.55006340041147,
     "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": [
-    {
-      "seconds": 2.0410720630025025,
-      "bytes": 6120614,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 1.370815718000813,
       "bytes": 5076360,
@@ -562,6 +556,12 @@
       "bytes": 991660,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.7213281089971133,
+      "bytes": 11058082,
+      "pages": 32,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -296,3 +296,4 @@
 {"step": 954600, "loss": 0.5306006073951721, "lr": 3e-05, "grad_norm": 0.3563700754565489, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139727347353, "tokens_per_second": 153349.1151525728, "data_wait_s": 1.4411001757252961e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 954700, "loss": 0.5207518935203552, "lr": 3e-05, "grad_norm": 0.3578637098034645, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139741967197, "tokens_per_second": 152458.56245650307, "data_wait_s": 7.56800000090152e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 954800, "loss": 0.5444334745407104, "lr": 3e-05, "grad_norm": 0.35233121796476646, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139756502665, "tokens_per_second": 150143.417514215, "data_wait_s": 1.5052999515319243e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 954900, "loss": 0.5239086151123047, "lr": 3e-05, "grad_norm": 0.34884229323315963, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139770930518, "tokens_per_second": 149787.61962677827, "data_wait_s": 1.3266002497402951e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:54d4ee4ec166b4d57cf65f836a04c3342f921dcddc8822d64bfae50bd89d8162
-size 635244
+oid sha256:3eb7a626c5bda3b28e11a68cd7c4f38d7a4cb3651d24ee1e263928c34513be0c
+size 637362