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": 77313.4346167502,
-      "required_pages_per_second_upper_estimate": 0.005348722798670468
+      "required_token_positions_per_second": 76958.14220201387,
+      "required_pages_per_second_upper_estimate": 0.005324142845027027
     },
     {
       "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": 9664.179327093774,
-      "required_pages_per_second_upper_estimate": 0.004046190073241118
+      "required_token_positions_per_second": 9619.767775251734,
+      "required_pages_per_second_upper_estimate": 0.004027595883903519
     },
     {
       "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": 19328.35865418755,
-      "required_pages_per_second_upper_estimate": 0.004025587384572162
+      "required_token_positions_per_second": 19239.535550503468,
+      "required_pages_per_second_upper_estimate": 0.004007087874497533
     },
     {
       "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": 28992.537981281323,
-      "required_pages_per_second_upper_estimate": 0.0017380583829355787
+      "required_token_positions_per_second": 28859.3033257552,
+      "required_pages_per_second_upper_estimate": 0.001730071168774327
     },
     {
       "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": 28992.537981281323,
-      "required_pages_per_second_upper_estimate": 0.0022540423033472687
+      "required_token_positions_per_second": 28859.3033257552,
+      "required_pages_per_second_upper_estimate": 0.002243683894917428
     },
     {
       "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": 28992.537981281323,
-      "required_pages_per_second_upper_estimate": 0.37188514746195306
+      "required_token_positions_per_second": 28859.3033257552,
+      "required_pages_per_second_upper_estimate": 0.3701761563570914
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.37363027887494127,
-    "producer_p95_seconds": 0.35255342700111214,
-    "gpu_update_seconds": 0.9536246884577885,
-    "producer_batches_per_second": 2.676442613299851,
-    "gpu_batches_per_second": 1.0486305693461127,
-    "producer_headroom": 2.55232175328322,
+    "producer_batch_seconds": 0.11916911312533784,
+    "producer_p95_seconds": 0.34206829399772687,
+    "gpu_update_seconds": 0.9580272845784816,
+    "producer_batches_per_second": 8.39143611774836,
+    "gpu_batches_per_second": 1.0438116075576969,
+    "producer_headroom": 8.039224757600255,
     "sustainable": true,
-    "buffer_seconds": 60.078355372840676,
+    "buffer_seconds": 60.35571892844434,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 406963.5478953458,
+    "generation_bytes_per_second": 404571.7195626911,
     "upload_bytes_per_second": 2943952.740836387,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.32490375180511,
+    "estimated_full_shards_per_hour": 86.81167307053137,
     "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": [

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -424,3 +424,4 @@
 {"step": 884900, "loss": 0.538706362247467, "lr": 3e-05, "grad_norm": 0.3125689198798968, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129524800048, "tokens_per_second": 152526.70202692781, "data_wait_s": 1.8385995645076036e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 885000, "loss": 0.4727434515953064, "lr": 3e-05, "grad_norm": 0.322338081895821, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129539339413, "tokens_per_second": 151282.6534431105, "data_wait_s": 1.4504003047477454e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 885100, "loss": 0.5561938285827637, "lr": 3e-05, "grad_norm": 0.320365593019778, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129553907620, "tokens_per_second": 139508.06889502337, "data_wait_s": 1.631199847906828e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 885200, "loss": 0.5439466834068298, "lr": 3e-05, "grad_norm": 0.3224920051910945, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129568451569, "tokens_per_second": 150826.9158274706, "data_wait_s": 8.47400224301964e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e81cbe7612c318e313bbecf3fa1451622590d3a9471bccaad944dec51289b37d
-size 909480
+oid sha256:1f608a5106340b128a571eb9e442293086f1ce88dbf83b1e09537f4334df4aff
+size 911598