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": 74593.6392598537,
-      "required_pages_per_second_upper_estimate": 0.006465587595193847
+      "required_token_positions_per_second": 74351.968090257,
+      "required_pages_per_second_upper_estimate": 0.0064446401507231845
     },
     {
       "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": 9324.204907481713,
-      "required_pages_per_second_upper_estimate": 0.0037195647468811683
+      "required_token_positions_per_second": 9293.996011282125,
+      "required_pages_per_second_upper_estimate": 0.0037075139665239047
     },
     {
       "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": 18648.409814963426,
-      "required_pages_per_second_upper_estimate": 0.004313981409888494
+      "required_token_positions_per_second": 18587.99202256425,
+      "required_pages_per_second_upper_estimate": 0.004300004816933775
     },
     {
       "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": 27972.61472244514,
-      "required_pages_per_second_upper_estimate": 0.0017033854756297456
+      "required_token_positions_per_second": 27881.988033846374,
+      "required_pages_per_second_upper_estimate": 0.0016978667857755687
     },
     {
       "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": 27972.61472244514,
-      "required_pages_per_second_upper_estimate": 0.002153409314487321
+      "required_token_positions_per_second": 27881.988033846374,
+      "required_pages_per_second_upper_estimate": 0.0021464326211282575
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27972.61472244514,
-      "required_pages_per_second_upper_estimate": 0.3473262565335826
+      "required_token_positions_per_second": 27881.988033846374,
+      "required_pages_per_second_upper_estimate": 0.3462009763692014
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.12515130437481048,
-    "producer_p95_seconds": 0.31810767099887016,
-    "gpu_update_seconds": 0.9883952670972631,
-    "producer_batches_per_second": 7.9903282270645875,
-    "gpu_batches_per_second": 1.0117409838847347,
-    "producer_headroom": 7.897602602184304,
+    "producer_batch_seconds": 0.34470340527343524,
+    "producer_p95_seconds": 0.33099639200008824,
+    "gpu_update_seconds": 0.9916079142720264,
+    "producer_batches_per_second": 2.901044737886335,
+    "gpu_batches_per_second": 1.0084631088630778,
+    "producer_headroom": 2.876698921745306,
     "sustainable": true,
-    "buffer_seconds": 60.29211129293305,
+    "buffer_seconds": 60.488082770593614,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 456233.19046689157,
-    "upload_bytes_per_second": 2799686.7873858884,
+    "generation_bytes_per_second": 430250.1575970408,
+    "upload_bytes_per_second": 2797455.874187011,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.89702210907994,
+    "estimated_full_shards_per_hour": 92.32166810925882,
     "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": 1.240897143001348,
-      "bytes": 7984887,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.0015397960014525,
       "bytes": 971215,
@@ -562,6 +556,12 @@
       "bytes": 1151300,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.3565457900003821,
+      "bytes": 7973344,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -240,3 +240,4 @@
 {"step": 219000, "loss": 0.635651670396328, "lr": 0.0003, "grad_norm": 0.1789175456113893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32054448077, "tokens_per_second": 149585.77841550292, "data_wait_s": 1.5058998542372137e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 219100, "loss": 0.6620151866227388, "lr": 0.0003, "grad_norm": 0.16267989680756995, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32069099202, "tokens_per_second": 148261.5626955664, "data_wait_s": 1.4432996977120638e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 219200, "loss": 0.5834460780024529, "lr": 0.0003, "grad_norm": 0.15104916600150237, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32083818204, "tokens_per_second": 149474.60381156084, "data_wait_s": 1.1145999451400712e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 219300, "loss": 0.5816761590540409, "lr": 0.0003, "grad_norm": 0.14476674569779652, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32098461872, "tokens_per_second": 148982.575581119, "data_wait_s": 1.609200262464583e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3a7bb707e68288e2363fb96e1403735fb0c94eb4f0eecdc487b551fb61c11765
-size 515853
+oid sha256:25b3b75187b0521e9b93acd59154a55dbacdf67b92a2df5ee409afaec24461e1
+size 517971