Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75671.18642372194,
-      "required_pages_per_second_upper_estimate": 0.006558986652339628
+      "required_token_positions_per_second": 75638.64878272531,
+      "required_pages_per_second_upper_estimate": 0.006556166372083937
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9458.898302965243,
-      "required_pages_per_second_upper_estimate": 0.003773295956185273
+      "required_token_positions_per_second": 9454.831097840664,
+      "required_pages_per_second_upper_estimate": 0.0037716734872509433
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18917.796605930485,
-      "required_pages_per_second_upper_estimate": 0.004376299302933129
+      "required_token_positions_per_second": 18909.66219568133,
+      "required_pages_per_second_upper_estimate": 0.004374417550282718
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28376.694908895726,
-      "required_pages_per_second_upper_estimate": 0.0017279918389396937
+      "required_token_positions_per_second": 28364.493293521995,
+      "required_pages_per_second_upper_estimate": 0.0017272488245803617
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28376.694908895726,
-      "required_pages_per_second_upper_estimate": 0.0021845165258058347
+      "required_token_positions_per_second": 28364.493293521995,
+      "required_pages_per_second_upper_estimate": 0.0021835772116781315
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28376.694908895726,
-      "required_pages_per_second_upper_estimate": 0.3523435800799102
+      "required_token_positions_per_second": 28364.493293521995,
+      "required_pages_per_second_upper_estimate": 0.3521920768531482
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.508925801078064,
-    "producer_p95_seconds": 0.2563514329995087,
-    "gpu_update_seconds": 0.9743206560441509,
-    "producer_batches_per_second": 1.9649229767515959,
-    "gpu_batches_per_second": 1.0263561526654994,
-    "producer_headroom": 1.9144650437848407,
+    "producer_batch_seconds": 0.11424014910933522,
+    "producer_p95_seconds": 0.27410942999995314,
+    "gpu_update_seconds": 0.9747397816661992,
+    "producer_batches_per_second": 8.75348997525323,
+    "gpu_batches_per_second": 1.0259148326650027,
+    "producer_headroom": 8.532374907295596,
     "sustainable": true,
-    "buffer_seconds": 60.407880674737356,
+    "buffer_seconds": 60.433866463304355,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 278001.2352467758,
-    "upload_bytes_per_second": 4060128.621004436,
+    "generation_bytes_per_second": 263816.6870277666,
+    "upload_bytes_per_second": 4033853.5570244105,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.65259354641395,
+    "estimated_full_shards_per_hour": 56.60891969799756,
     "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": 1.0493596160004017,
-      "bytes": 7456331,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 0.9663659530006043,
       "bytes": 5966016,
@@ -562,6 +556,12 @@
       "bytes": 5874669,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.0266657669999404,
+      "bytes": 4655742,
+      "pages": 12,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -129,3 +129,4 @@
 {"step": 327900, "loss": 0.5953261367976665, "lr": 0.0003, "grad_norm": 0.14521236999780726, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47995493904, "tokens_per_second": 148830.59430890298, "data_wait_s": 1.3767999917035922e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 328000, "loss": 0.5767523758113384, "lr": 0.0003, "grad_norm": 0.1545782758922978, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48010136677, "tokens_per_second": 131453.26330401003, "data_wait_s": 1.0034998922492377e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 328100, "loss": 0.553047776222229, "lr": 0.0003, "grad_norm": 0.17460492853849696, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48024741114, "tokens_per_second": 131876.6009304648, "data_wait_s": 1.2071999663021415e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 328200, "loss": 0.6228276044130325, "lr": 0.0003, "grad_norm": 0.16822864349379343, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48039404181, "tokens_per_second": 149127.94836256246, "data_wait_s": 1.2841999705415219e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:782ef75193834cc8755c1ef3979ff0582fcc17ed37ad7f703da2455d9ac25d97
-size 279189
+oid sha256:2b8b9b83db6e785ddbe5b413d80848ec40c3de7e812456f298ea6854c762a4d3
+size 281307