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": 75755.20948634214,
-      "required_pages_per_second_upper_estimate": 0.005240921168568406
+      "required_token_positions_per_second": 74330.9312234634,
+      "required_pages_per_second_upper_estimate": 0.005142386293561575
     },
     {
       "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": 9469.401185792767,
-      "required_pages_per_second_upper_estimate": 0.0039646405329085
+      "required_token_positions_per_second": 9291.366402932925,
+      "required_pages_per_second_upper_estimate": 0.0038901010871141136
     },
     {
       "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": 18938.802371585534,
-      "required_pages_per_second_upper_estimate": 0.0039444530841961836
+      "required_token_positions_per_second": 18582.73280586585,
+      "required_pages_per_second_upper_estimate": 0.0038702931838426836
     },
     {
       "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": 28408.2035573783,
-      "required_pages_per_second_upper_estimate": 0.001703028426449596
+      "required_token_positions_per_second": 27874.099208798776,
+      "required_pages_per_second_upper_estimate": 0.0016710096862823732
     },
     {
       "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": 28408.2035573783,
-      "required_pages_per_second_upper_estimate": 0.0022086128721043106
+      "required_token_positions_per_second": 27874.099208798776,
+      "required_pages_per_second_upper_estimate": 0.002167088608279003
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28408.2035573783,
-      "required_pages_per_second_upper_estimate": 0.3643899328815472
+      "required_token_positions_per_second": 27874.099208798776,
+      "required_pages_per_second_upper_estimate": 0.3575390157745382
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.10442269464033416,
-    "producer_p95_seconds": 0.31385164999664994,
-    "gpu_update_seconds": 0.9732399989375302,
-    "producer_batches_per_second": 9.576462314483708,
-    "gpu_batches_per_second": 1.0274957883889722,
-    "producer_headroom": 9.320196172773423,
+    "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,
     "sustainable": true,
-    "buffer_seconds": 60.340879934126875,
+    "buffer_seconds": 60.50520188532687,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 350310.05828677124,
-    "upload_bytes_per_second": 5445753.153991307,
+    "generation_bytes_per_second": 350159.98056434555,
+    "upload_bytes_per_second": 5373573.357883692,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.16838370754579,
+    "estimated_full_shards_per_hour": 75.1361805219438,
     "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": 1.2589823859998432,
-      "bytes": 12076255,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.1457863180003187,
       "bytes": 6106429,
@@ -562,6 +556,12 @@
       "bytes": 7185173,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 1.1344652970001334,
+      "bytes": 5709671,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -336,3 +336,4 @@
 {"step": 473800, "loss": 0.6174167394638062, "lr": 0.0003, "grad_norm": 0.17022402488085814, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69349458590, "tokens_per_second": 150756.6135311059, "data_wait_s": 1.1961994459852576e-05, "gpu_peak_gb": 89.37899398803711}
 {"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}

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 sha256:38487986ebc73642d81bee753362447b5a4bfeffcd87ae59d4757606bc02a7ae
-size 720747
+oid sha256:cea5607c832aa8166b4ee5d53a6ed8203661d15350fe51ea8221ab99034b598a
+size 722865