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": 75784.22107044638,
-      "required_pages_per_second_upper_estimate": 0.006568784209034009
+      "required_token_positions_per_second": 75761.30885298371,
+      "required_pages_per_second_upper_estimate": 0.006566798236094829
     },
     {
       "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": 9473.027633805797,
-      "required_pages_per_second_upper_estimate": 0.003778932357509892
+      "required_token_positions_per_second": 9470.163606622964,
+      "required_pages_per_second_upper_estimate": 0.0037777898542456375
     },
     {
       "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": 18946.055267611595,
-      "required_pages_per_second_upper_estimate": 0.004382836446977595
+      "required_token_positions_per_second": 18940.327213245928,
+      "required_pages_per_second_upper_estimate": 0.004381511362410412
     },
     {
       "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": 28419.082901417394,
-      "required_pages_per_second_upper_estimate": 0.001730573045291654
+      "required_token_positions_per_second": 28410.490819868894,
+      "required_pages_per_second_upper_estimate": 0.001730049832604518
     },
     {
       "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": 28419.082901417394,
-      "required_pages_per_second_upper_estimate": 0.002187779670807626
+      "required_token_positions_per_second": 28410.490819868894,
+      "required_pages_per_second_upper_estimate": 0.0021871182285856184
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28419.082901417394,
-      "required_pages_per_second_upper_estimate": 0.35286989708354416
+      "required_token_positions_per_second": 28410.490819868894,
+      "required_pages_per_second_upper_estimate": 0.35276321218655887
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.1275102241871764,
-    "producer_p95_seconds": 0.3155896110001777,
-    "gpu_update_seconds": 0.972867424888685,
-    "producer_batches_per_second": 7.842508366482578,
-    "gpu_batches_per_second": 1.0278892831820527,
-    "producer_headroom": 7.629720919167873,
+    "producer_batch_seconds": 0.6132215871484448,
+    "producer_p95_seconds": 0.35302726599911693,
+    "gpu_update_seconds": 0.9731616456504284,
+    "producer_batches_per_second": 1.6307318935886161,
+    "gpu_batches_per_second": 1.0275785163436375,
+    "producer_headroom": 1.586965733179337,
     "sustainable": true,
-    "buffer_seconds": 60.31778034309848,
+    "buffer_seconds": 60.33602203032656,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 374164.3168560717,
-    "upload_bytes_per_second": 4803578.213540358,
+    "generation_bytes_per_second": 371255.4969490323,
+    "upload_bytes_per_second": 4904741.672189561,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.28695229779828,
+    "estimated_full_shards_per_hour": 79.66278725960947,
     "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": 1.070391503002611,
-      "bytes": 1123754,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.18286236700078,
       "bytes": 5852988,
@@ -562,6 +556,12 @@
       "bytes": 2620953,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.4711359560023993,
+      "bytes": 11859463,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -314,3 +314,4 @@
 {"step": 268500, "loss": 0.5589021369814873, "lr": 0.0003, "grad_norm": 0.1745483632741585, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39300244150, "tokens_per_second": 149560.46126671965, "data_wait_s": 1.2019001587759703e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 268600, "loss": 0.5550219640135765, "lr": 0.0003, "grad_norm": 0.1480452006594306, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39314917054, "tokens_per_second": 149776.29686238288, "data_wait_s": 1.7141002899734303e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 268700, "loss": 0.5741091929376125, "lr": 0.0003, "grad_norm": 0.1482366127425559, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39329559086, "tokens_per_second": 149792.38781241022, "data_wait_s": 1.2163000064902008e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 268800, "loss": 0.5637525208294392, "lr": 0.0003, "grad_norm": 0.15044971965244122, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39344183795, "tokens_per_second": 134925.3938784452, "data_wait_s": 1.3763998140348122e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4f0f69aee6d26530da99e73f37ab985506a658250c84246061da716bb49d8c7b
-size 674151
+oid sha256:6e8f09c6aafb7225104513dd01907d55e79329de6b09be5dc04e512360066306
+size 676269