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": 77521.55102975943,
-      "required_pages_per_second_upper_estimate": 0.005363120775019055
+      "required_token_positions_per_second": 77352.66846981032,
+      "required_pages_per_second_upper_estimate": 0.0053514370876602485
     },
     {
       "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": 9690.193878719929,
-      "required_pages_per_second_upper_estimate": 0.004057081822761377
+      "required_token_positions_per_second": 9669.08355872629,
+      "required_pages_per_second_upper_estimate": 0.004048243372613651
     },
     {
       "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": 19380.387757439858,
-      "required_pages_per_second_upper_estimate": 0.004036423674679895
+      "required_token_positions_per_second": 19338.16711745258,
+      "required_pages_per_second_upper_estimate": 0.004027630228803697
     },
     {
       "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": 29070.58163615979,
-      "required_pages_per_second_upper_estimate": 0.0017427369808797816
+      "required_token_positions_per_second": 29007.250676178868,
+      "required_pages_per_second_upper_estimate": 0.0017389403865297021
     },
     {
       "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": 29070.58163615979,
-      "required_pages_per_second_upper_estimate": 0.0022601098542363103
+      "required_token_positions_per_second": 29007.250676178868,
+      "required_pages_per_second_upper_estimate": 0.002255186150661247
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 29070.58163615979,
-      "required_pages_per_second_upper_estimate": 0.37288620766998615
+      "required_token_positions_per_second": 29007.250676178868,
+      "required_pages_per_second_upper_estimate": 0.3720738661148378
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08494271275004195,
-    "producer_p95_seconds": 0.17340131099990685,
-    "gpu_update_seconds": 0.9510645623137346,
-    "producer_batches_per_second": 11.77264026100351,
-    "gpu_batches_per_second": 1.051453328854159,
-    "producer_headroom": 11.196540957108354,
+    "producer_batch_seconds": 0.07082616519529239,
+    "producer_p95_seconds": 0.17755827800010593,
+    "gpu_update_seconds": 0.9531410028701858,
+    "producer_batches_per_second": 14.119075870374344,
+    "gpu_batches_per_second": 1.0491627125354046,
+    "producer_headroom": 13.457470134688844,
     "sustainable": true,
-    "buffer_seconds": 60.868131988079014,
+    "buffer_seconds": 60.04788318082171,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 316254.0899718103,
-    "upload_bytes_per_second": 5422405.077498226,
+    "generation_bytes_per_second": 282806.5325754368,
+    "upload_bytes_per_second": 5468571.436015223,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.86076568952305,
+    "estimated_full_shards_per_hour": 60.68369849154785,
     "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": [
@@ -550,6 +550,12 @@
       "bytes": 8185407,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.3622697179998795,
+      "bytes": 10825600,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -73,3 +73,4 @@
 {"step": 564400, "loss": 0.5654087662696838, "lr": 0.0003, "grad_norm": 0.165348709344066, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82610859212, "tokens_per_second": 154824.6561223058, "data_wait_s": 7.89299974712776e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 564500, "loss": 0.6321799159049988, "lr": 0.0003, "grad_norm": 0.16649485057799607, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82625466448, "tokens_per_second": 152091.64661745608, "data_wait_s": 1.189700014947448e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 564600, "loss": 0.5881126523017883, "lr": 0.0003, "grad_norm": 0.15395510416711883, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82640203619, "tokens_per_second": 149025.17739549733, "data_wait_s": 1.7798999579099473e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 564700, "loss": 0.5635547041893005, "lr": 0.0003, "grad_norm": 0.17389888051443375, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82654803122, "tokens_per_second": 152585.3997248335, "data_wait_s": 1.554000027681468e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c2774476fa04ac695c13759878785a3c052325771319078971fbc304718551e7
-size 159798
+oid sha256:808b99a0792af8e1b796bb5a9ded3df3dcbb8e58320925b9f757c782e532722f
+size 161916