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": 77461.66477695967,
-      "required_pages_per_second_upper_estimate": 0.005358977705095117
+      "required_token_positions_per_second": 77038.97900703353,
+      "required_pages_per_second_upper_estimate": 0.005329735322765517
     },
     {
       "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": 9682.70809711996,
-      "required_pages_per_second_upper_estimate": 0.004053947682326365
+      "required_token_positions_per_second": 9629.872375879191,
+      "required_pages_per_second_upper_estimate": 0.0040318264691781795
     },
     {
       "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": 19365.41619423992,
-      "required_pages_per_second_upper_estimate": 0.004033305492892021
+      "required_token_positions_per_second": 19259.744751758382,
+      "required_pages_per_second_upper_estimate": 0.004011296918166455
     },
     {
       "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": 29048.124291359876,
-      "required_pages_per_second_upper_estimate": 0.0017413906973494067
+      "required_token_positions_per_second": 28889.617127637575,
+      "required_pages_per_second_upper_estimate": 0.0017318884349106802
     },
     {
       "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": 29048.124291359876,
-      "required_pages_per_second_upper_estimate": 0.0022583638944575378
+      "required_token_positions_per_second": 28889.617127637575,
+      "required_pages_per_second_upper_estimate": 0.0022460406596774597
     },
     {
       "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": 29048.124291359876,
-      "required_pages_per_second_upper_estimate": 0.372598148963711
+      "required_token_positions_per_second": 28889.617127637575,
+      "required_pages_per_second_upper_estimate": 0.3705649892592139
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11270888579690563,
-    "producer_p95_seconds": 0.3178241200002958,
-    "gpu_update_seconds": 0.9517998381817606,
-    "producer_batches_per_second": 8.872414920345655,
-    "gpu_batches_per_second": 1.0506410695659678,
-    "producer_headroom": 8.444763085466432,
+    "producer_batch_seconds": 0.3958365550077474,
+    "producer_p95_seconds": 0.31999602299765684,
+    "gpu_update_seconds": 0.9570220289818321,
+    "producer_batches_per_second": 2.5262952280403406,
+    "gpu_batches_per_second": 1.0449080268966138,
+    "producer_headroom": 2.417720184946287,
     "sustainable": true,
-    "buffer_seconds": 60.91518964363268,
+    "buffer_seconds": 60.29238782585542,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 361245.4276955363,
-    "upload_bytes_per_second": 4879072.873709683,
+    "generation_bytes_per_second": 334364.4587982968,
+    "upload_bytes_per_second": 4900837.938239857,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.5148594202954,
+    "estimated_full_shards_per_hour": 71.74682925187757,
     "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": 1.141177229997993,
-      "bytes": 6001543,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 0.9294679570011795,
       "bytes": 2588442,
@@ -562,6 +556,12 @@
       "bytes": 856694,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.058450859003642,
+      "bytes": 7478371,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -347,3 +347,4 @@
 {"step": 591800, "loss": 0.6143639087677002, "lr": 0.0003, "grad_norm": 0.19243285126151685, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86621653055, "tokens_per_second": 153704.5501025626, "data_wait_s": 8.636998245492578e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 591900, "loss": 0.5977855324745178, "lr": 0.0003, "grad_norm": 0.16993496243179834, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86636358255, "tokens_per_second": 152570.45149698132, "data_wait_s": 1.22580022434704e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 592000, "loss": 0.6148668527603149, "lr": 0.0003, "grad_norm": 0.19262962241393378, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86650986318, "tokens_per_second": 152141.1724714603, "data_wait_s": 1.084299583453685e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 592100, "loss": 0.5621565580368042, "lr": 0.0003, "grad_norm": 0.16573701637922625, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86665577589, "tokens_per_second": 151326.54263661295, "data_wait_s": 1.1258998711127788e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bf0439006d5b99a973d09cc0637b422061f20e1ada25e15f9a668d924b3c4854
-size 744828
+oid sha256:2bee38cc89662cd0b9f4079faaeb007a8e57e05092642b77a759a771c1a8cd16
+size 746946