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": 74330.9312234634,
-      "required_pages_per_second_upper_estimate": 0.005142386293561575
+      "required_token_positions_per_second": 76524.62854031172,
+      "required_pages_per_second_upper_estimate": 0.005294151364020193
     },
     {
       "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": 9291.366402932925,
-      "required_pages_per_second_upper_estimate": 0.0038901010871141136
+      "required_token_positions_per_second": 9565.578567538965,
+      "required_pages_per_second_upper_estimate": 0.004004907994233518
     },
     {
       "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": 18582.73280586585,
-      "required_pages_per_second_upper_estimate": 0.0038702931838426836
+      "required_token_positions_per_second": 19131.15713507793,
+      "required_pages_per_second_upper_estimate": 0.00398451550869541
     },
     {
       "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": 27874.099208798776,
-      "required_pages_per_second_upper_estimate": 0.0016710096862823732
+      "required_token_positions_per_second": 28696.735702616897,
+      "required_pages_per_second_upper_estimate": 0.0017203254879935737
     },
     {
       "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": 27874.099208798776,
-      "required_pages_per_second_upper_estimate": 0.002167088608279003
+      "required_token_positions_per_second": 28696.735702616897,
+      "required_pages_per_second_upper_estimate": 0.002231044977277829
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 27874.099208798776,
-      "required_pages_per_second_upper_estimate": 0.3575390157745382
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28696.735702616897,
+      "required_pages_per_second_upper_estimate": 0.3680909134389874
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.12757407840564383,
-    "producer_p95_seconds": 0.3389192370013916,
-    "gpu_update_seconds": 0.9918885554971618,
-    "producer_batches_per_second": 7.83858298251097,
-    "gpu_batches_per_second": 1.0081777780960206,
-    "producer_headroom": 7.775000751667441,
+    "producer_batch_seconds": 0.14130945435090325,
+    "producer_p95_seconds": 0.3107515619994956,
+    "gpu_update_seconds": 0.9634545296898956,
+    "producer_batches_per_second": 7.076667336898594,
+    "gpu_batches_per_second": 1.037931702206919,
+    "producer_headroom": 6.818047200843481,
     "sustainable": true,
-    "buffer_seconds": 60.50520188532687,
+    "buffer_seconds": 60.69763537046342,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 350159.98056434555,
-    "upload_bytes_per_second": 5373573.357883692,
+    "generation_bytes_per_second": 424871.41366858274,
+    "upload_bytes_per_second": 5383268.544429008,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.1361805219438,
+    "estimated_full_shards_per_hour": 91.16751487296212,
     "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": 1.1457863180003187,
-      "bytes": 6106429,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.9636820249979792,
       "bytes": 6954429,
@@ -562,6 +556,12 @@
       "bytes": 5709671,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.2757203309956822,
+      "bytes": 7569977,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -337,3 +337,4 @@
 {"step": 473900, "loss": 0.5944867134094238, "lr": 0.0003, "grad_norm": 0.1693777827876962, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69364032372, "tokens_per_second": 151665.13956717157, "data_wait_s": 9.452000085730106e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 474000, "loss": 0.5799126029014587, "lr": 0.0003, "grad_norm": 0.15906559848364973, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69378778683, "tokens_per_second": 153242.5071553906, "data_wait_s": 1.0492003639228642e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 474100, "loss": 0.6103756427764893, "lr": 0.0003, "grad_norm": 0.1460226722104052, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69393242883, "tokens_per_second": 149687.91178289353, "data_wait_s": 1.7116006347350776e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 474200, "loss": 0.5817875862121582, "lr": 0.0003, "grad_norm": 0.15582655146374855, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69407863053, "tokens_per_second": 152170.59249969205, "data_wait_s": 1.9254999642726034e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cea5607c832aa8166b4ee5d53a6ed8203661d15350fe51ea8221ab99034b598a
-size 722865
+oid sha256:26e350abb300632fe1c60e4f2b1c1a25bbfafbe501dda0972d2833978e1fafee
+size 724983