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": 76363.10932999066,
-      "required_pages_per_second_upper_estimate": 0.0052829770902739805
+      "required_token_positions_per_second": 76610.2230337651,
+      "required_pages_per_second_upper_estimate": 0.0053000729896838855
     },
     {
       "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": 9545.388666248833,
-      "required_pages_per_second_upper_estimate": 0.003996454904176422
+      "required_token_positions_per_second": 9576.277879220637,
+      "required_pages_per_second_upper_estimate": 0.004009387572607599
     },
     {
       "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": 19090.777332497666,
-      "required_pages_per_second_upper_estimate": 0.003976105460704945
+      "required_token_positions_per_second": 19152.555758441275,
+      "required_pages_per_second_upper_estimate": 0.003988972277622347
     },
     {
       "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": 28636.165998746503,
-      "required_pages_per_second_upper_estimate": 0.0017166944267311263
+      "required_token_positions_per_second": 28728.83363766191,
+      "required_pages_per_second_upper_estimate": 0.0017222497101888133
     },
     {
       "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": 28636.165998746503,
-      "required_pages_per_second_upper_estimate": 0.0022263359492198775
+      "required_token_positions_per_second": 28728.83363766191,
+      "required_pages_per_second_upper_estimate": 0.0022335404505436835
     },
     {
       "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": 28636.165998746503,
-      "required_pages_per_second_upper_estimate": 0.36731399031241907
+      "required_token_positions_per_second": 28728.83363766191,
+      "required_pages_per_second_upper_estimate": 0.36850263128566735
     }
   ],
   "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": 0.429775696867182,
-    "producer_p95_seconds": 0.33563627200055635,
-    "gpu_update_seconds": 0.9654923777579109,
-    "producer_batches_per_second": 2.326795133576481,
-    "gpu_batches_per_second": 1.035740957709292,
-    "producer_headroom": 2.246502966072293,
+    "producer_batch_seconds": 0.09386425778927787,
+    "producer_p95_seconds": 0.33516216999851167,
+    "gpu_update_seconds": 0.9623780884635359,
+    "producer_batches_per_second": 10.653682493766334,
+    "gpu_batches_per_second": 1.039092651825156,
+    "producer_headroom": 10.252870593448282,
     "sustainable": true,
-    "buffer_seconds": 60.82601979874838,
+    "buffer_seconds": 60.62981957320276,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 295136.64226745936,
+    "generation_bytes_per_second": 314974.6053637319,
     "upload_bytes_per_second": 5158322.939816339,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.329453001192434,
+    "estimated_full_shards_per_hour": 67.58621807750671,
     "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-v1/1791399463488568880-054ba646/train.jsonl

@@ -382,3 +382,4 @@
 {"step": 923200, "loss": 0.5117206573486328, "lr": 3e-05, "grad_norm": 0.3306566202676778, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135130219494, "tokens_per_second": 153538.5503625293, "data_wait_s": 1.2519994925241917e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 923300, "loss": 0.5540512800216675, "lr": 3e-05, "grad_norm": 0.33279744847035647, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135144911724, "tokens_per_second": 152360.5292352392, "data_wait_s": 1.515400072094053e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 923400, "loss": 0.5590641498565674, "lr": 3e-05, "grad_norm": 0.36065657908464965, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135159570767, "tokens_per_second": 152866.02151184296, "data_wait_s": 1.4614997780881822e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 923500, "loss": 0.5171303749084473, "lr": 3e-05, "grad_norm": 0.33135059818747514, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135174161331, "tokens_per_second": 152485.88817410546, "data_wait_s": 1.572800101712346e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:aed733375d134f748f3494c711fccdb61a0ecc50561eecdb93799e2b42c9c46e
-size 818958
+oid sha256:667053085da4739090928805bd1644738dbcfb5b389e4d8f805c5faacc04d952
+size 821076