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": 75608.98886691632,
-      "required_pages_per_second_upper_estimate": 0.006553595525753983
+      "required_token_positions_per_second": 75669.16504189603,
+      "required_pages_per_second_upper_estimate": 0.006558811444086103
     },
     {
       "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": 9451.12360836454,
-      "required_pages_per_second_upper_estimate": 0.0037701945142670098
+      "required_token_positions_per_second": 9458.645630237004,
+      "required_pages_per_second_upper_estimate": 0.003773195161256185
     },
     {
       "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": 18902.24721672908,
-      "required_pages_per_second_upper_estimate": 0.004372702225401288
+      "required_token_positions_per_second": 18917.29126047401,
+      "required_pages_per_second_upper_estimate": 0.0043761824001555495
     },
     {
       "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": 28353.370825093618,
-      "required_pages_per_second_upper_estimate": 0.0017265715246081556
+      "required_token_positions_per_second": 28375.936890711015,
+      "required_pages_per_second_upper_estimate": 0.0017279456796092587
     },
     {
       "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": 28353.370825093618,
-      "required_pages_per_second_upper_estimate": 0.0021827209732695517
+      "required_token_positions_per_second": 28375.936890711015,
+      "required_pages_per_second_upper_estimate": 0.0021844581714676467
     },
     {
       "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": 28353.370825093618,
-      "required_pages_per_second_upper_estimate": 0.352053973019775
+      "required_token_positions_per_second": 28375.936890711015,
+      "required_pages_per_second_upper_estimate": 0.3523341680309797
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.07593554861722396,
-    "producer_p95_seconds": 0.3145949589998054,
-    "gpu_update_seconds": 0.9751221528669408,
-    "producer_batches_per_second": 13.169062688159425,
-    "gpu_batches_per_second": 1.0255125443103885,
-    "producer_headroom": 12.84144475971772,
+    "producer_batch_seconds": 0.09104133510928136,
+    "producer_p95_seconds": 0.31450611599939293,
+    "gpu_update_seconds": 0.9743466834764033,
+    "producer_batches_per_second": 10.984021695196486,
+    "gpu_batches_per_second": 1.0263287359198139,
+    "producer_headroom": 10.702245109947556,
     "sustainable": true,
-    "buffer_seconds": 60.45757347775033,
+    "buffer_seconds": 60.409494375537,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 316202.6023194876,
-    "upload_bytes_per_second": 6056738.078214915,
+    "generation_bytes_per_second": 366516.02421782457,
+    "upload_bytes_per_second": 6015171.334181287,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.8497176379058,
+    "estimated_full_shards_per_hour": 78.64580674077084,
     "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": 1.0901677820002078,
-      "bytes": 6115749,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 1.5552335219981615,
       "bytes": 12252293,
@@ -562,6 +556,12 @@
       "bytes": 6038268,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 0.7658507999985886,
+      "bytes": 1224241,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -179,3 +179,4 @@
 {"step": 102900, "loss": 0.5999361556023359, "lr": 0.0003, "grad_norm": 0.13805758448796915, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15057865595, "tokens_per_second": 148841.67114829342, "data_wait_s": 1.5591998817399144e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 103000, "loss": 0.638515941798687, "lr": 0.0003, "grad_norm": 0.13897537475620358, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15072547623, "tokens_per_second": 149250.89938169898, "data_wait_s": 1.4072000340092927e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 103100, "loss": 0.6555536761879921, "lr": 0.0003, "grad_norm": 0.1502817850151909, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15087338923, "tokens_per_second": 150393.0267746047, "data_wait_s": 1.6738998965593055e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 103200, "loss": 0.5946326497942209, "lr": 0.0003, "grad_norm": 0.14209048301726152, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15102059084, "tokens_per_second": 149550.96910927029, "data_wait_s": 1.8296999769518152e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e28555b2a86e78048c178f59d9402a8d5a1dd0495a4c2dd428726355f7e18150
-size 385872
+oid sha256:72e55dde8d7f0cae019ddf0dd0a6aa222c55e1fb6fb78fa8169b1a7767ca6ad9
+size 387990