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": 75738.33364701176,
-      "required_pages_per_second_upper_estimate": 0.006564806803471308
+      "required_token_positions_per_second": 75867.71998504823,
+      "required_pages_per_second_upper_estimate": 0.006576021683325633
     },
     {
       "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": 9467.29170587647,
-      "required_pages_per_second_upper_estimate": 0.0037766442100991187
+      "required_token_positions_per_second": 9483.464998131029,
+      "required_pages_per_second_upper_estimate": 0.00378309597819173
     },
     {
       "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": 18934.58341175294,
-      "required_pages_per_second_upper_estimate": 0.004380182635022462
+      "required_token_positions_per_second": 18966.929996262057,
+      "required_pages_per_second_upper_estimate": 0.004387665448068203
     },
     {
       "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": 28401.875117629414,
-      "required_pages_per_second_upper_estimate": 0.00172952518154123
+      "required_token_positions_per_second": 28450.394994393086,
+      "required_pages_per_second_upper_estimate": 0.001732479787472018
     },
     {
       "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": 28401.875117629414,
-      "required_pages_per_second_upper_estimate": 0.0021864549679774326
+      "required_token_positions_per_second": 28450.394994393086,
+      "required_pages_per_second_upper_estimate": 0.0021901901624023154
     },
     {
       "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": 28401.875117629414,
-      "required_pages_per_second_upper_estimate": 0.3526562339996451
+      "required_token_positions_per_second": 28450.394994393086,
+      "required_pages_per_second_upper_estimate": 0.3532586884834683
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.05316818744530849,
-    "producer_p95_seconds": 0.1766920609989029,
-    "gpu_update_seconds": 0.9734568540102666,
-    "producer_batches_per_second": 18.808239438830803,
-    "gpu_batches_per_second": 1.027266895168888,
-    "producer_headroom": 18.30900959359606,
+    "producer_batch_seconds": 0.10839925453126398,
+    "producer_p95_seconds": 0.1904043829999864,
+    "gpu_update_seconds": 0.9717967010809088,
+    "producer_batches_per_second": 9.225155692482968,
+    "gpu_batches_per_second": 1.029021809693037,
+    "producer_headroom": 8.964975868912715,
     "sustainable": true,
-    "buffer_seconds": 60.35432494863653,
+    "buffer_seconds": 60.251395467016344,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 361396.0339606937,
-    "upload_bytes_per_second": 5754149.303409202,
+    "generation_bytes_per_second": 373588.8639750047,
+    "upload_bytes_per_second": 5844458.057408248,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.54717601886375,
+    "estimated_full_shards_per_hour": 80.16347350537878,
     "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": [
@@ -304,6 +304,12 @@
       "bytes": 5962190,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.9607173089989374,
+      "bytes": 7750938,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -25,3 +25,4 @@
 {"step": 282700, "loss": 0.589450684376061, "lr": 0.0003, "grad_norm": 0.15634303942668046, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41379596319, "tokens_per_second": 151108.18336983526, "data_wait_s": 1.4613999155699275e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 282800, "loss": 0.653716541826725, "lr": 0.0003, "grad_norm": 0.1598144832022321, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41394340143, "tokens_per_second": 150733.54300563177, "data_wait_s": 1.3720000424655154e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 282900, "loss": 0.5886063631623983, "lr": 0.0003, "grad_norm": 0.1679207484828719, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41408818163, "tokens_per_second": 147980.88940932546, "data_wait_s": 8.70500116434414e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 283000, "loss": 0.5849144067615271, "lr": 0.0003, "grad_norm": 0.1710524655987325, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41423324239, "tokens_per_second": 127116.93380241522, "data_wait_s": 9.403000149177387e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:182c91b1119310cc7440291cb8884eb38c844e5f066daa0f7706f0bc517c63b8
-size 57351
+oid sha256:ffd4bce07bd233dcbe3537efe49799d3451c6ff3a5579d2d6b37f7cac46f0568
+size 59469