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": 77108.25695910935,
-      "required_pages_per_second_upper_estimate": 0.005334528132237118
+      "required_token_positions_per_second": 77259.24261973142,
+      "required_pages_per_second_upper_estimate": 0.005344973670573952
     },
     {
       "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": 9638.532119888669,
-      "required_pages_per_second_upper_estimate": 0.004035452123158929
+      "required_token_positions_per_second": 9657.405327466427,
+      "required_pages_per_second_upper_estimate": 0.004043353941054346
     },
     {
       "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": 19277.064239777337,
-      "required_pages_per_second_upper_estimate": 0.00401490411077519
+      "required_token_positions_per_second": 19314.810654932855,
+      "required_pages_per_second_upper_estimate": 0.004022765693612176
     },
     {
       "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": 28915.596359666008,
-      "required_pages_per_second_upper_estimate": 0.0017334458502027941
+      "required_token_positions_per_second": 28972.215982399284,
+      "required_pages_per_second_upper_estimate": 0.001736840110132496
     },
     {
       "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": 28915.596359666008,
-      "required_pages_per_second_upper_estimate": 0.002248060430697105
+      "required_token_positions_per_second": 28972.215982399284,
+      "required_pages_per_second_upper_estimate": 0.0022524623573212167
     },
     {
       "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": 28915.596359666008,
-      "required_pages_per_second_upper_estimate": 0.37089822295334857
+      "required_token_positions_per_second": 28972.215982399284,
+      "required_pages_per_second_upper_estimate": 0.37162447868035664
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.18330724660157216,
-    "producer_p95_seconds": 0.16104374099995766,
-    "gpu_update_seconds": 0.9561621920606777,
-    "producer_batches_per_second": 5.455321699166384,
-    "gpu_batches_per_second": 1.045847669258753,
-    "producer_headroom": 5.2161723542711105,
+    "producer_batch_seconds": 0.09169713060936857,
+    "producer_p95_seconds": 0.1631191790002049,
+    "gpu_update_seconds": 0.9542935899965764,
+    "producer_batches_per_second": 10.905466652604627,
+    "gpu_batches_per_second": 1.0478955433448816,
+    "producer_headroom": 10.407016922502017,
     "sustainable": true,
-    "buffer_seconds": 60.23821809982269,
+    "buffer_seconds": 60.12049616978432,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 449063.6832705555,
-    "upload_bytes_per_second": 5522888.685692098,
+    "generation_bytes_per_second": 448657.23369413894,
+    "upload_bytes_per_second": 5567276.432905611,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.35861276233196,
+    "estimated_full_shards_per_hour": 96.27139814489485,
     "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": [
@@ -400,6 +400,12 @@
       "bytes": 6309981,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.1507453160002115,
+      "bytes": 8207384,
+      "pages": 33,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -42,3 +42,4 @@
 {"step": 561300, "loss": 0.5954387187957764, "lr": 0.0003, "grad_norm": 0.18554076959041468, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82156827772, "tokens_per_second": 152629.9568855695, "data_wait_s": 1.4752999959455337e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 561400, "loss": 0.581609308719635, "lr": 0.0003, "grad_norm": 0.1562346964315421, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82171487848, "tokens_per_second": 152818.35751960153, "data_wait_s": 1.3908000255469233e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 561500, "loss": 0.5722002387046814, "lr": 0.0003, "grad_norm": 0.1494081318000216, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82186091670, "tokens_per_second": 151706.1245322584, "data_wait_s": 1.5481999980693217e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 561600, "loss": 0.592818021774292, "lr": 0.0003, "grad_norm": 0.15642986707509363, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82200627359, "tokens_per_second": 150529.44197882366, "data_wait_s": 1.7706999642541632e-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:d8bde9086576520a1ae382d5aa39fcb7dbc4c5175070f1f24385777333f285ae
-size 94140
+oid sha256:bf69f9f7c505ab1f407ebfff7573673b96589ce11df003689cd219c105d0af19
+size 96258