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": 76559.18284179673,
-      "required_pages_per_second_upper_estimate": 0.005296541910773952
+      "required_token_positions_per_second": 76342.26255933823,
+      "required_pages_per_second_upper_estimate": 0.005281534862309072
     },
     {
       "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": 9569.897855224592,
-      "required_pages_per_second_upper_estimate": 0.004006716389790506
+      "required_token_positions_per_second": 9542.782819917278,
+      "required_pages_per_second_upper_estimate": 0.003995363890733659
     },
     {
       "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": 19139.795710449183,
-      "required_pages_per_second_upper_estimate": 0.003986314696130689
+      "required_token_positions_per_second": 19085.565639834556,
+      "required_pages_per_second_upper_estimate": 0.0039750200025647974
     },
     {
       "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": 28709.693565673777,
-      "required_pages_per_second_upper_estimate": 0.0017211022920983223
+      "required_token_positions_per_second": 28628.348459751833,
+      "required_pages_per_second_upper_estimate": 0.0017162257772050866
     },
     {
       "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": 28709.693565673777,
-      "required_pages_per_second_upper_estimate": 0.0022320523941348857
+      "required_token_positions_per_second": 28628.348459751833,
+      "required_pages_per_second_upper_estimate": 0.002225728169948762
     },
     {
       "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": 28709.693565673777,
-      "required_pages_per_second_upper_estimate": 0.36825712299321167
+      "required_token_positions_per_second": 28628.348459751833,
+      "required_pages_per_second_upper_estimate": 0.3672137153160148
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10149786353926515,
-    "producer_p95_seconds": 0.2606014819975826,
-    "gpu_update_seconds": 0.9630196831169535,
-    "producer_batches_per_second": 9.85242413120492,
-    "gpu_batches_per_second": 1.0384003749158628,
-    "producer_headroom": 9.488078364766787,
+    "producer_batch_seconds": 0.09203488360151368,
+    "producer_p95_seconds": 0.2504416920019139,
+    "gpu_update_seconds": 0.9657560246225838,
+    "producer_batches_per_second": 10.865445371015314,
+    "gpu_batches_per_second": 1.035458205286163,
+    "producer_headroom": 10.493369327265604,
     "sustainable": true,
-    "buffer_seconds": 60.67024003636807,
+    "buffer_seconds": 60.84262955122278,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 417883.37695913325,
-    "upload_bytes_per_second": 3930080.089339655,
+    "generation_bytes_per_second": 514709.2665774475,
+    "upload_bytes_per_second": 3936235.4823358366,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.66804486828326,
+    "estimated_full_shards_per_hour": 110.44462678902214,
     "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": 1.9822834630012949,
-      "bytes": 7671850,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.43599132899908,
       "bytes": 5917528,
@@ -562,6 +556,12 @@
       "bytes": 2589704,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.4273169570005848,
+      "bytes": 6151606,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -229,3 +229,4 @@
 {"step": 822900, "loss": 0.5381624698638916, "lr": 3e-05, "grad_norm": 0.2898037839632764, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120449300301, "tokens_per_second": 149929.63971339003, "data_wait_s": 1.5158002497628331e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 823000, "loss": 0.5252871513366699, "lr": 3e-05, "grad_norm": 0.26089176789618596, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120463834068, "tokens_per_second": 150460.31595709507, "data_wait_s": 1.3984001270728186e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 823100, "loss": 0.5351876616477966, "lr": 3e-05, "grad_norm": 0.2871865533724164, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120478645494, "tokens_per_second": 152926.0526024939, "data_wait_s": 1.365699790767394e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 823200, "loss": 0.5542900562286377, "lr": 3e-05, "grad_norm": 0.2848713946231784, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120493358351, "tokens_per_second": 152233.3260610851, "data_wait_s": 9.142000635620207e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5e11a7bc652c13be3b040ebe8683ad20c9ee52a7b9e5f2baedeb12bbb959bf2c
-size 492555
+oid sha256:796c850d075d7093268a704ea38f298be3255743de4923196e1d9a7edb9ce313
+size 494673