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": 76625.00793601619,
-      "required_pages_per_second_upper_estimate": 0.005301095843527315
+      "required_token_positions_per_second": 77118.43064890416,
+      "required_pages_per_second_upper_estimate": 0.005335231971703335
     },
     {
       "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": 9578.125992002024,
-      "required_pages_per_second_upper_estimate": 0.004010161338836182
+      "required_token_positions_per_second": 9639.80383111302,
+      "required_pages_per_second_upper_estimate": 0.004035984562092215
     },
     {
       "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": 19156.251984004048,
-      "required_pages_per_second_upper_estimate": 0.00398974210393105
+      "required_token_positions_per_second": 19279.60766222604,
+      "required_pages_per_second_upper_estimate": 0.004015433838596694
     },
     {
       "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": 28734.37797600607,
-      "required_pages_per_second_upper_estimate": 0.001722582084806831
+      "required_token_positions_per_second": 28919.41149333906,
+      "required_pages_per_second_upper_estimate": 0.0017336745616411218
     },
     {
       "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": 28734.37797600607,
-      "required_pages_per_second_upper_estimate": 0.002233971498462975
+      "required_token_positions_per_second": 28919.41149333906,
+      "required_pages_per_second_upper_estimate": 0.0022483570405591852
     },
     {
       "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": 28734.37797600607,
-      "required_pages_per_second_upper_estimate": 0.3685737481048995
+      "required_token_positions_per_second": 28919.41149333906,
+      "required_pages_per_second_upper_estimate": 0.3709471593917351
     }
   ],
   "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": 0.14464302983623156,
-    "producer_p95_seconds": 0.323587832004705,
-    "gpu_update_seconds": 0.9621923962678703,
-    "producer_batches_per_second": 6.913571992596013,
-    "gpu_batches_per_second": 1.039293184896053,
-    "producer_headroom": 6.652186402326393,
+    "producer_batch_seconds": 0.10556999719591431,
+    "producer_p95_seconds": 0.3063197880037478,
+    "gpu_update_seconds": 0.9560360523369605,
+    "producer_batches_per_second": 9.472388240611805,
+    "gpu_batches_per_second": 1.0459856587579233,
+    "producer_headroom": 9.055944659757557,
     "sustainable": true,
-    "buffer_seconds": 60.61812096487583,
+    "buffer_seconds": 60.23027129722851,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 339104.2948460365,
+    "generation_bytes_per_second": 325745.4943308062,
     "upload_bytes_per_second": 4770646.14230003,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.7638877299864,
+    "estimated_full_shards_per_hour": 69.89740011637821,
     "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-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -383,3 +383,4 @@
 {"step": 798300, "loss": 0.5799460411071777, "lr": 3e-05, "grad_norm": 0.2847245988695983, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116847986781, "tokens_per_second": 150896.43604142455, "data_wait_s": 1.353699917672202e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 798400, "loss": 0.5292727947235107, "lr": 3e-05, "grad_norm": 0.26238272934409546, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116862567200, "tokens_per_second": 151617.72200572313, "data_wait_s": 8.610004442743957e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 798500, "loss": 0.5403978824615479, "lr": 3e-05, "grad_norm": 0.27010619785582285, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116877223179, "tokens_per_second": 153099.01971590056, "data_wait_s": 7.811999239493161e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 798600, "loss": 0.5269435048103333, "lr": 3e-05, "grad_norm": 0.27219380500039625, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116892025692, "tokens_per_second": 154238.1250671306, "data_wait_s": 1.463400258217007e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:018d16aa883c78a740093df927c9da2927b92906b314dc1bcaa86a8d619a6aeb
-size 821076
+oid sha256:ae10429780b821cf375b01f4c880026b520b087ecb41e5d009e725ec61ca560c
+size 823194