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": 74975.1252179315,
-      "required_pages_per_second_upper_estimate": 0.00649865383117271
+      "required_token_positions_per_second": 75255.1226721041,
+      "required_pages_per_second_upper_estimate": 0.006522923300853331
     },
     {
       "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": 9371.890652241438,
-      "required_pages_per_second_upper_estimate": 0.003738587303431242
+      "required_token_positions_per_second": 9406.890334013013,
+      "required_pages_per_second_upper_estimate": 0.0037525491997817987
     },
     {
       "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": 18743.781304482876,
-      "required_pages_per_second_upper_estimate": 0.004336043925507931
+      "required_token_positions_per_second": 18813.780668026026,
+      "required_pages_per_second_upper_estimate": 0.004352237046316917
     },
     {
       "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": 28115.671956724316,
-      "required_pages_per_second_upper_estimate": 0.0017120969106340458
+      "required_token_positions_per_second": 28220.67100203904,
+      "required_pages_per_second_upper_estimate": 0.0017184908016062964
     },
     {
       "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": 28115.671956724316,
-      "required_pages_per_second_upper_estimate": 0.002164422256389902
+      "required_token_positions_per_second": 28220.67100203904,
+      "required_pages_per_second_upper_estimate": 0.0021725053735541897
     },
     {
       "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": 28115.671956724316,
-      "required_pages_per_second_upper_estimate": 0.3491025486015659
+      "required_token_positions_per_second": 28220.67100203904,
+      "required_pages_per_second_upper_estimate": 0.3504062853351756
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.2284291896484234,
-    "producer_p95_seconds": 0.24707863299954624,
-    "gpu_update_seconds": 0.9833661469146404,
-    "producer_batches_per_second": 4.377724237165598,
-    "gpu_batches_per_second": 1.01691521834217,
-    "producer_headroom": 4.304905815356367,
+    "producer_batch_seconds": 0.13972836331254257,
+    "producer_p95_seconds": 0.26482249600121577,
+    "gpu_update_seconds": 0.9797073924288455,
+    "producer_batches_per_second": 7.156743099918899,
+    "gpu_batches_per_second": 1.0207129268677313,
+    "producer_headroom": 7.0115141207046765,
     "sustainable": true,
-    "buffer_seconds": 60.968701108707705,
+    "buffer_seconds": 60.74185833058842,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 432318.93406882114,
-    "upload_bytes_per_second": 5250166.155069888,
+    "generation_bytes_per_second": 427844.63215868216,
+    "upload_bytes_per_second": 5289281.255193045,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.76557938145137,
+    "estimated_full_shards_per_hour": 91.8054983479533,
     "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": 1.7142034060007063,
-      "bytes": 6133193,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0039940289998412,
       "bytes": 2541534,
@@ -562,6 +556,12 @@
       "bytes": 6179293,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.0649322700010089,
+      "bytes": 5823579,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -134,3 +134,4 @@
 {"step": 208400, "loss": 0.5508072338998318, "lr": 0.0003, "grad_norm": 0.1494632014841731, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30501775990, "tokens_per_second": 146016.4629212084, "data_wait_s": 1.8898999769589864e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 208500, "loss": 0.6042758673429489, "lr": 0.0003, "grad_norm": 0.14450466836031053, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30516426206, "tokens_per_second": 148336.9974279627, "data_wait_s": 2.116799987561535e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 208600, "loss": 0.6167097222059965, "lr": 0.0003, "grad_norm": 0.14051531890274305, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30531144904, "tokens_per_second": 149534.9880224501, "data_wait_s": 2.0044000848429278e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 208700, "loss": 0.5843361467123032, "lr": 0.0003, "grad_norm": 0.1446629908582026, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30545964578, "tokens_per_second": 142373.67929929346, "data_wait_s": 1.7577000107849017e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fc285daf5d36fbd15707c9cc1d1743c287177e1c5702575f0be566913e2c2133
-size 289779
+oid sha256:04051d6a2a3b2d81708d5e8a66e301789db0d28b40d68a3a37397e030d7a35eb
+size 291897