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": 76011.44851677855,
-      "required_pages_per_second_upper_estimate": 0.006588479708179384
+      "required_token_positions_per_second": 76031.68052557671,
+      "required_pages_per_second_upper_estimate": 0.006590233367424451
     },
     {
       "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": 9501.431064597318,
-      "required_pages_per_second_upper_estimate": 0.003790262910721764
+      "required_token_positions_per_second": 9503.960065697089,
+      "required_pages_per_second_upper_estimate": 0.003791271767072399
     },
     {
       "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": 19002.862129194637,
-      "required_pages_per_second_upper_estimate": 0.004395977714638214
+      "required_token_positions_per_second": 19007.920131394178,
+      "required_pages_per_second_upper_estimate": 0.004397147794429279
     },
     {
       "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": 28504.293193791957,
-      "required_pages_per_second_upper_estimate": 0.0017357619050334114
+      "required_token_positions_per_second": 28511.880197091265,
+      "required_pages_per_second_upper_estimate": 0.0017362239137284094
     },
     {
       "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": 28504.293193791957,
-      "required_pages_per_second_upper_estimate": 0.0021943393949918058
+      "required_token_positions_per_second": 28511.880197091265,
+      "required_pages_per_second_upper_estimate": 0.0021949234635079556
     },
     {
       "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": 28504.293193791957,
-      "required_pages_per_second_upper_estimate": 0.35392792373433274
+      "required_token_positions_per_second": 28511.880197091265,
+      "required_pages_per_second_upper_estimate": 0.35402212892324353
     }
   ],
   "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": 0.09074777631252573,
-    "producer_p95_seconds": 0.34983883099630475,
-    "gpu_update_seconds": 0.9699591500841547,
-    "producer_batches_per_second": 11.019553763566678,
-    "gpu_batches_per_second": 1.0309712526689798,
-    "producer_headroom": 10.688517002815784,
+    "producer_batch_seconds": 0.12217652149246305,
+    "producer_p95_seconds": 0.37709177199576516,
+    "gpu_update_seconds": 0.9697010442271922,
+    "producer_batches_per_second": 8.184878631216302,
+    "gpu_batches_per_second": 1.0312456668508125,
+    "producer_headroom": 7.936885355563279,
     "sustainable": true,
-    "buffer_seconds": 60.13746730521759,
+    "buffer_seconds": 60.12146474208591,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 296807.9644553606,
-    "upload_bytes_per_second": 5272264.636593416,
+    "generation_bytes_per_second": 295024.98705112346,
+    "upload_bytes_per_second": 5252190.819026769,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.68807983632673,
+    "estimated_full_shards_per_hour": 63.305494390967155,
     "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": 1.2613436630017532,
-      "bytes": 7588531,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 0.8843855299965071,
       "bytes": 5949979,
@@ -562,6 +556,12 @@
       "bytes": 2314805,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.2841032439973787,
+      "bytes": 6126814,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -357,3 +357,4 @@
 {"step": 272800, "loss": 0.6262881923466921, "lr": 0.0003, "grad_norm": 0.16025266180820172, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39929169684, "tokens_per_second": 148093.05563199896, "data_wait_s": 1.3905999367125332e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 272900, "loss": 0.5818833615630865, "lr": 0.0003, "grad_norm": 0.1676216423082919, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39943781282, "tokens_per_second": 149702.61720655984, "data_wait_s": 1.1689997336361557e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 273000, "loss": 0.5703057590872049, "lr": 0.0003, "grad_norm": 0.14665260047803771, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39958325289, "tokens_per_second": 148488.69058170656, "data_wait_s": 1.4484001439996064e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 273100, "loss": 0.5965057983994484, "lr": 0.0003, "grad_norm": 0.16545406222730363, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39973025471, "tokens_per_second": 150156.61626166187, "data_wait_s": 1.3949000276625156e-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 sha256:806cf9017ba603eaa1b80bc7ee0bba9da374b2d5a9a83b7f8770ebd973f0dc79
-size 766008
+oid sha256:abd7fd909a65006638a68593fc346ee75788c4062a24fcaf011148c9b3fb4652
+size 768126