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": 76318.46267625397,
-      "required_pages_per_second_upper_estimate": 0.005279888331173967
+      "required_token_positions_per_second": 76379.1046660074,
+      "required_pages_per_second_upper_estimate": 0.005284083684733911
     },
     {
       "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": 9539.807834531746,
-      "required_pages_per_second_upper_estimate": 0.003994118326477496
+      "required_token_positions_per_second": 9547.388083250926,
+      "required_pages_per_second_upper_estimate": 0.003997292018322623
     },
     {
       "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": 19079.615669063493,
-      "required_pages_per_second_upper_estimate": 0.003973780780564466
+      "required_token_positions_per_second": 19094.77616650185,
+      "required_pages_per_second_upper_estimate": 0.003976938312371672
     },
     {
       "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": 28619.42350359524,
-      "required_pages_per_second_upper_estimate": 0.0017156907397111178
+      "required_token_positions_per_second": 28642.164249752776,
+      "required_pages_per_second_upper_estimate": 0.0017170540127201532
     },
     {
       "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": 28619.42350359524,
-      "required_pages_per_second_upper_estimate": 0.0022250342938643203
+      "required_token_positions_per_second": 28642.164249752776,
+      "required_pages_per_second_upper_estimate": 0.002226802286852097
     },
     {
       "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": 28619.42350359524,
-      "required_pages_per_second_upper_estimate": 0.36709923556130936
+      "required_token_positions_per_second": 28642.164249752776,
+      "required_pages_per_second_upper_estimate": 0.3673909294359074
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.07104133738280893,
-    "producer_p95_seconds": 0.17874336299973947,
-    "gpu_update_seconds": 0.9660571952655438,
-    "producer_batches_per_second": 14.076311579150913,
-    "gpu_batches_per_second": 1.035135398712212,
-    "producer_headroom": 13.598522083838429,
+    "producer_batch_seconds": 0.05519175929687492,
+    "producer_p95_seconds": 0.18207385200003046,
+    "gpu_update_seconds": 0.9652901840418237,
+    "producer_batches_per_second": 18.11864692736153,
+    "gpu_batches_per_second": 1.035957908338859,
+    "producer_headroom": 17.489752027101634,
     "sustainable": true,
-    "buffer_seconds": 60.86160330172926,
+    "buffer_seconds": 60.81328159463489,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 353044.891872456,
+    "generation_bytes_per_second": 391943.1751564111,
     "upload_bytes_per_second": 4812616.01237047,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.75521533136616,
+    "estimated_full_shards_per_hour": 84.10188141841172,
     "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": [

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -70,3 +70,4 @@
 {"step": 727000, "loss": 0.5028425455093384, "lr": 5e-05, "grad_norm": 0.21711830382886493, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106412387595, "tokens_per_second": 151712.87865452308, "data_wait_s": 1.2137000339862425e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 727100, "loss": 0.5169442296028137, "lr": 5e-05, "grad_norm": 0.22224307388557282, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106427141789, "tokens_per_second": 152708.3475259535, "data_wait_s": 1.4548000763170421e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 727200, "loss": 0.50984787940979, "lr": 5e-05, "grad_norm": 0.21736101496838955, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106441840632, "tokens_per_second": 151554.5667339757, "data_wait_s": 1.0786000530060846e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 727300, "loss": 0.5268535017967224, "lr": 5e-05, "grad_norm": 0.23538101980784507, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106456330822, "tokens_per_second": 129230.42199775409, "data_wait_s": 1.3830000170855783e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:94cb01171a2e148632e47b57eda1af42068a7225d0ed2a8d8ecced46c1e63525
-size 153444
+oid sha256:69931690a67537d487ef61ad7ada8519cd29d6d1212bfe60ae48fc740c392e19
+size 155562