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": 77176.89841787542,
-      "required_pages_per_second_upper_estimate": 0.005339276907624682
+      "required_token_positions_per_second": 76709.18041178661,
+      "required_pages_per_second_upper_estimate": 0.005306919090708165
     },
     {
       "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": 9647.112302234427,
-      "required_pages_per_second_upper_estimate": 0.004039044466332516
+      "required_token_positions_per_second": 9588.647551473327,
+      "required_pages_per_second_upper_estimate": 0.0040145664960716705
     },
     {
       "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": 19294.224604468855,
-      "required_pages_per_second_upper_estimate": 0.004018478162191183
+      "required_token_positions_per_second": 19177.295102946653,
+      "required_pages_per_second_upper_estimate": 0.003994124830662429
     },
     {
       "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": 28941.33690670328,
-      "required_pages_per_second_upper_estimate": 0.001734988956694139
+      "required_token_positions_per_second": 28765.94265441998,
+      "required_pages_per_second_upper_estimate": 0.0017244743390812717
     },
     {
       "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": 28941.33690670328,
-      "required_pages_per_second_upper_estimate": 0.002250061645008552
+      "required_token_positions_per_second": 28765.94265441998,
+      "required_pages_per_second_upper_estimate": 0.0022364255133713092
     },
     {
       "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": 28941.33690670328,
-      "required_pages_per_second_upper_estimate": 0.3712283950526966
+      "required_token_positions_per_second": 28765.94265441998,
+      "required_pages_per_second_upper_estimate": 0.368978625908082
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,39 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09608320560147376,
-    "producer_p95_seconds": 0.27539721799985273,
-    "gpu_update_seconds": 0.9553117773766794,
-    "producer_batches_per_second": 10.407646099441353,
-    "gpu_batches_per_second": 1.0467786786278674,
-    "producer_headroom": 9.942546893564783,
+    "producer_batch_seconds": 0.40209721585921443,
+    "producer_p95_seconds": 0.2900204580000718,
+    "gpu_update_seconds": 0.9611365889221709,
+    "producer_batches_per_second": 2.486960766100226,
+    "gpu_batches_per_second": 1.0404348471650746,
+    "producer_headroom": 2.3903089875128405,
     "sustainable": true,
-    "buffer_seconds": 60.1846419747308,
+    "buffer_seconds": 60.551605102096765,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 397695.1736651185,
-    "upload_bytes_per_second": 4823252.873992993,
+    "generation_bytes_per_second": 350166.00328952575,
+    "upload_bytes_per_second": 4834978.016490366,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.33612639870803,
+    "estimated_full_shards_per_hour": 75.13747285856562,
     "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": 1.369231013002718,
-      "bytes": 10621268,
-      "pages": 30,
-      "shards": 3
-    },
-    {
-      "seconds": 0.9135342390000005,
-      "bytes": 1211398,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.2419597409971175,
       "bytes": 13297063,
@@ -562,6 +550,18 @@
       "bytes": 5852982,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.0356132939996314,
+      "bytes": 7513178,
+      "pages": 37,
+      "shards": 3
+    },
+    {
+      "seconds": 1.2883466929997667,
+      "bytes": 5484455,
+      "pages": 22,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -298,3 +298,4 @@
 {"step": 714800, "loss": 0.552908182144165, "lr": 8e-05, "grad_norm": 0.2087740568057489, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104626872512, "tokens_per_second": 153083.12809174953, "data_wait_s": 9.734001650940627e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 714900, "loss": 0.5387517213821411, "lr": 8e-05, "grad_norm": 0.21496947351464413, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104641426324, "tokens_per_second": 151266.63435850193, "data_wait_s": 1.8170001567341387e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 715000, "loss": 0.49898889660835266, "lr": 8e-05, "grad_norm": 0.23847586120249772, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104656097809, "tokens_per_second": 152420.1536742034, "data_wait_s": 1.3430999388219789e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 715100, "loss": 0.5293847918510437, "lr": 8e-05, "grad_norm": 0.2190585582182101, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104670736248, "tokens_per_second": 134843.6941909461, "data_wait_s": 6.782000127714127e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2ade695ea281a30c06e6ccff8cb235a5dc43e28c8fec8e20c1a6c3d6f9541b33
-size 640263
+oid sha256:7859b6d84c0e766c41937f021966fde4ae56e1fc1f96cb7788d72a428c145484
+size 642381