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": 75536.22278310962,
-      "required_pages_per_second_upper_estimate": 0.006547288346033572
+      "required_token_positions_per_second": 75466.96730172266,
+      "required_pages_per_second_upper_estimate": 0.0065412854564333624
     },
     {
       "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": 9442.027847888703,
-      "required_pages_per_second_upper_estimate": 0.0037665660794194604
+      "required_token_positions_per_second": 9433.370912715332,
+      "required_pages_per_second_upper_estimate": 0.0037631126985460876
     },
     {
       "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": 18884.055695777406,
-      "required_pages_per_second_upper_estimate": 0.004368493937188952
+      "required_token_positions_per_second": 18866.741825430665,
+      "required_pages_per_second_upper_estimate": 0.004364488677997944
     },
     {
       "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": 28326.083543666107,
-      "required_pages_per_second_upper_estimate": 0.0017249098723345214
+      "required_token_positions_per_second": 28300.112738146,
+      "required_pages_per_second_upper_estimate": 0.0017233283865366329
     },
     {
       "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": 28326.083543666107,
-      "required_pages_per_second_upper_estimate": 0.0021806203228092328
+      "required_token_positions_per_second": 28300.112738146,
+      "required_pages_per_second_upper_estimate": 0.002178621018308504
     },
     {
       "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": 28326.083543666107,
-      "required_pages_per_second_upper_estimate": 0.3517151563091015
+      "required_token_positions_per_second": 28300.112738146,
+      "required_pages_per_second_upper_estimate": 0.3513926858232365
     }
   ],
   "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": 0.3666902746721519,
-    "producer_p95_seconds": 0.31879714200113085,
-    "gpu_update_seconds": 0.9760615143769944,
-    "producer_batches_per_second": 2.727097141842864,
-    "gpu_batches_per_second": 1.0245255911337567,
-    "producer_headroom": 2.6618145661203187,
+    "producer_batch_seconds": 0.5883339246094295,
+    "producer_p95_seconds": 0.3155310139991343,
+    "gpu_update_seconds": 0.9769572388569672,
+    "producer_batches_per_second": 1.6997150056642725,
+    "gpu_batches_per_second": 1.0235862535498408,
+    "producer_headroom": 1.6605488787775218,
     "sustainable": true,
-    "buffer_seconds": 60.515813891373654,
+    "buffer_seconds": 60.57134880913196,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 394153.70287763636,
-    "upload_bytes_per_second": 3946292.6944979886,
+    "generation_bytes_per_second": 366823.35255731456,
+    "upload_bytes_per_second": 3986869.82462545,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.57620920893496,
+    "estimated_full_shards_per_hour": 78.7117522481878,
     "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": 0.945500746000107,
-      "bytes": 979107,
-      "pages": 29,
-      "shards": 1
-    },
     {
       "seconds": 1.0416768409995711,
       "bytes": 11876824,
@@ -562,6 +556,12 @@
       "bytes": 5784149,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.3641967180010397,
+      "bytes": 7009240,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -230,3 +230,4 @@
 {"step": 138000, "loss": 0.6163253784179688, "lr": 0.0003, "grad_norm": 0.13840160992919742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20196193884, "tokens_per_second": 151910.37238622457, "data_wait_s": 8.898001397028565e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 138100, "loss": 0.618504099547863, "lr": 0.0003, "grad_norm": 0.17327833131624185, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20210930534, "tokens_per_second": 149147.4022133526, "data_wait_s": 1.811800029827282e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 138200, "loss": 0.6266972068697214, "lr": 0.0003, "grad_norm": 0.13924538543768078, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20225549777, "tokens_per_second": 149880.93232815288, "data_wait_s": 2.473600034136325e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 138300, "loss": 0.5591120338067412, "lr": 0.0003, "grad_norm": 0.12904902226672182, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20239986214, "tokens_per_second": 145486.039391103, "data_wait_s": 1.326400160905905e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:60e34cbc0aa19e0bd8214bdbb2173a81182ab58f3fe730a803713cad6dd69ad1
-size 494673
+oid sha256:73cb62a3e588149433cbdcabc8e1862ba131bb062df3bc91078f9455c86300b2
+size 496791