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": 75653.41075888592,
-      "required_pages_per_second_upper_estimate": 0.006557445902763652
+      "required_token_positions_per_second": 75748.15161257412,
+      "required_pages_per_second_upper_estimate": 0.006565657799842846
     },
     {
       "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": 9456.67634486074,
-      "required_pages_per_second_upper_estimate": 0.0037724095838761527
+      "required_token_positions_per_second": 9468.518951571765,
+      "required_pages_per_second_upper_estimate": 0.0037771337767559298
     },
     {
       "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": 18913.35268972148,
-      "required_pages_per_second_upper_estimate": 0.0043752712811284305
+      "required_token_positions_per_second": 18937.03790314353,
+      "required_pages_per_second_upper_estimate": 0.0043807504384081025
     },
     {
       "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": 28370.02903458222,
-      "required_pages_per_second_upper_estimate": 0.0017275859221671408
+      "required_token_positions_per_second": 28405.556854715294,
+      "required_pages_per_second_upper_estimate": 0.0017297493800131785
     },
     {
       "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": 28370.02903458222,
-      "required_pages_per_second_upper_estimate": 0.00218400336834886
+      "required_token_positions_per_second": 28405.556854715294,
+      "required_pages_per_second_upper_estimate": 0.0021867383982900044
     },
     {
       "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": 28370.02903458222,
-      "required_pages_per_second_upper_estimate": 0.3522608122301827
+      "required_token_positions_per_second": 28405.556854715294,
+      "required_pages_per_second_upper_estimate": 0.3527019488522703
     }
   ],
   "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": 0.07875426600003266,
+    "producer_batch_seconds": 0.11656077062494319,
     "producer_p95_seconds": 0.29636224299974856,
-    "gpu_update_seconds": 0.9745495842213595,
-    "producer_batches_per_second": 12.697724844512999,
-    "gpu_batches_per_second": 1.0261150547808964,
-    "producer_headroom": 12.37456246777737,
+    "gpu_update_seconds": 0.9733306810850448,
+    "producer_batches_per_second": 8.579215757055119,
+    "gpu_batches_per_second": 1.0274000598493669,
+    "producer_headroom": 8.350413915990007,
     "sustainable": true,
-    "buffer_seconds": 60.42207422172429,
+    "buffer_seconds": 60.34650222727278,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 490155.8800234727,
-    "upload_bytes_per_second": 4710759.968791011,
+    "generation_bytes_per_second": 512873.7689111375,
+    "upload_bytes_per_second": 4674834.667993678,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.17604160812508,
+    "estimated_full_shards_per_hour": 110.05077171803086,
     "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": 0.9742800499989244,
-      "bytes": 772407,
-      "pages": 25,
-      "shards": 1
-    },
     {
       "seconds": 1.406286874000216,
       "bytes": 18261180,
@@ -562,6 +556,12 @@
       "bytes": 6137294,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.6848288789988146,
+      "bytes": 823777,
+      "pages": 25,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -247,3 +247,4 @@
 {"step": 64700, "loss": 0.6381540410220623, "lr": 0.0003, "grad_norm": 0.15739835779292288, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9469111092, "tokens_per_second": 151236.28346579967, "data_wait_s": 8.705999789526686e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 64800, "loss": 0.6068667303770781, "lr": 0.0003, "grad_norm": 0.12981488414858786, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9483709383, "tokens_per_second": 148194.77878463193, "data_wait_s": 8.858998626237735e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 64900, "loss": 0.6011261399835348, "lr": 0.0003, "grad_norm": 0.13736492854896148, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9498208077, "tokens_per_second": 148070.96434491608, "data_wait_s": 1.1184001778019592e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 65000, "loss": 0.6491995714604855, "lr": 0.0003, "grad_norm": 0.12935625010639407, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9512775355, "tokens_per_second": 148919.8180950165, "data_wait_s": 1.1274001735728234e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:258ee811d776c0d7daf73af8fa2be59cc6dbefce51d0a19e886f47901576ecfc
-size 530679
+oid sha256:aa134008d5b67f3d1e300e4cbe470ed00e9c11c6ba378d3a0c37a2aaeb5c4323
+size 532797