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": 77169.21850422093,
-      "required_pages_per_second_upper_estimate": 0.005338745593378209
+      "required_token_positions_per_second": 77213.67016974345,
+      "required_pages_per_second_upper_estimate": 0.005341820862741132
     },
     {
       "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": 9646.152313027616,
-      "required_pages_per_second_upper_estimate": 0.004038642538898479
+      "required_token_positions_per_second": 9651.708771217931,
+      "required_pages_per_second_upper_estimate": 0.004040968911910722
     },
     {
       "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": 19292.304626055233,
-      "required_pages_per_second_upper_estimate": 0.004018078281320861
+      "required_token_positions_per_second": 19303.417542435862,
+      "required_pages_per_second_upper_estimate": 0.004020392808735633
     },
     {
       "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": 28938.456939082847,
-      "required_pages_per_second_upper_estimate": 0.00173481630703275
+      "required_token_positions_per_second": 28955.126313653793,
+      "required_pages_per_second_upper_estimate": 0.001735815610585617
     },
     {
       "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": 28938.456939082847,
-      "required_pages_per_second_upper_estimate": 0.002249837740193702
+      "required_token_positions_per_second": 28955.126313653793,
+      "required_pages_per_second_upper_estimate": 0.0022511337107457632
     },
     {
       "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": 28938.456939082847,
-      "required_pages_per_second_upper_estimate": 0.37119145392033
+      "required_token_positions_per_second": 28955.126313653793,
+      "required_pages_per_second_upper_estimate": 0.37140527075914614
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.20431219131251765,
-    "producer_p95_seconds": 0.1442799550000018,
-    "gpu_update_seconds": 0.9554068504136438,
-    "producer_batches_per_second": 4.894470533431808,
-    "gpu_batches_per_second": 1.0466745131323367,
-    "producer_headroom": 4.676210676788472,
+    "producer_batch_seconds": 0.06715492870314321,
+    "producer_p95_seconds": 0.15040887099985412,
+    "gpu_update_seconds": 0.9548568257138834,
+    "producer_batches_per_second": 14.890939791187577,
+    "gpu_batches_per_second": 1.0472774274325012,
+    "producer_headroom": 14.218715500909928,
     "sustainable": true,
-    "buffer_seconds": 60.19063157605956,
+    "buffer_seconds": 60.155980019974656,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 406748.89767974796,
-    "upload_bytes_per_second": 5415057.947084561,
+    "generation_bytes_per_second": 406325.31236660416,
+    "upload_bytes_per_second": 5339466.466482653,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.2788448123391,
+    "estimated_full_shards_per_hour": 87.18795326469987,
     "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": [
@@ -262,6 +262,12 @@
       "bytes": 10919751,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.3756260560003284,
+      "bytes": 6013324,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -18,3 +18,4 @@
 {"step": 442000, "loss": 0.564111053943634, "lr": 0.0003, "grad_norm": 0.18659389085639008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64695562707, "tokens_per_second": 152270.7116784014, "data_wait_s": 1.3111000043863896e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 442100, "loss": 0.5817098021507263, "lr": 0.0003, "grad_norm": 0.16716236427496123, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64710338440, "tokens_per_second": 153906.83277553265, "data_wait_s": 1.580200023454381e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 442200, "loss": 0.5698598623275757, "lr": 0.0003, "grad_norm": 0.16258972882849176, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64725016724, "tokens_per_second": 152293.02992094893, "data_wait_s": 1.3707000107388012e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 442300, "loss": 0.5957845449447632, "lr": 0.0003, "grad_norm": 0.15658732256581287, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64739600490, "tokens_per_second": 151319.5431445865, "data_wait_s": 1.513600000180304e-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 sha256:6ae7551a54615c16c6849b7ef17b87bf52e8bf4e31a31251a400d8694948286e
-size 42525
+oid sha256:2f6bbd463cf74cbd6a2386a48cd6730f4086d12149a400dccfb36d535c9411ca
+size 44643