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": 76398.54854285614,
-      "required_pages_per_second_upper_estimate": 0.005285428857250317
+      "required_token_positions_per_second": 76868.45296201363,
+      "required_pages_per_second_upper_estimate": 0.005317937935296103
     },
     {
       "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": 9549.818567857017,
-      "required_pages_per_second_upper_estimate": 0.003998309611472904
+      "required_token_positions_per_second": 9608.556620251704,
+      "required_pages_per_second_upper_estimate": 0.004022902007420544
     },
     {
       "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": 19099.637135714034,
-      "required_pages_per_second_upper_estimate": 0.003977950724066192
+      "required_token_positions_per_second": 19217.113240503408,
+      "required_pages_per_second_upper_estimate": 0.0040024178986406
     },
     {
       "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": 28649.455703571053,
-      "required_pages_per_second_upper_estimate": 0.0017174911242431538
+      "required_token_positions_per_second": 28825.669860755108,
+      "required_pages_per_second_upper_estimate": 0.001728054893902899
     },
     {
       "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": 28649.455703571053,
-      "required_pages_per_second_upper_estimate": 0.0022273691653147526
+      "required_token_positions_per_second": 28825.669860755108,
+      "required_pages_per_second_upper_estimate": 0.002241069040951649
     },
     {
       "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": 28649.455703571053,
-      "required_pages_per_second_upper_estimate": 0.3674844563765351
+      "required_token_positions_per_second": 28825.669860755108,
+      "required_pages_per_second_upper_estimate": 0.3697447423808713
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08817315757823962,
-    "producer_p95_seconds": 0.31775968999863835,
-    "gpu_update_seconds": 0.9650445120517167,
-    "producer_batches_per_second": 11.341320050976504,
-    "gpu_batches_per_second": 1.0362216327969855,
-    "producer_headroom": 10.944878674616971,
+    "producer_batch_seconds": 0.12200571312490638,
+    "producer_p95_seconds": 0.296693004998815,
+    "gpu_update_seconds": 0.9591450999597253,
+    "producer_batches_per_second": 8.196337486066945,
+    "gpu_batches_per_second": 1.0425951193849503,
+    "producer_headroom": 7.861476937377323,
     "sustainable": true,
-    "buffer_seconds": 60.79780425925815,
+    "buffer_seconds": 60.42614129746269,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 331869.42466010136,
-    "upload_bytes_per_second": 4687209.403315414,
+    "generation_bytes_per_second": 329726.85517263395,
+    "upload_bytes_per_second": 4677588.592484768,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.21145300724297,
+    "estimated_full_shards_per_hour": 70.75170747169746,
     "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": 1.4296780900003796,
-      "bytes": 6878310,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.173846097997739,
       "bytes": 6578162,
@@ -562,6 +556,12 @@
       "bytes": 9600274,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.406910452999,
+      "bytes": 5890041,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -315,3 +315,4 @@
 {"step": 874000, "loss": 0.5015284419059753, "lr": 3e-05, "grad_norm": 0.32555418062398894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127929404004, "tokens_per_second": 153155.30740216115, "data_wait_s": 1.6776000848039985e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 874100, "loss": 0.49958536028862, "lr": 3e-05, "grad_norm": 0.3206786860098578, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127943824634, "tokens_per_second": 149745.2708233356, "data_wait_s": 1.2484000762924552e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 874200, "loss": 0.5261496901512146, "lr": 3e-05, "grad_norm": 0.3134866176176676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127958325044, "tokens_per_second": 151313.49266729451, "data_wait_s": 1.093399987439625e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 874300, "loss": 0.5033455491065979, "lr": 3e-05, "grad_norm": 0.30888795372106576, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127972816590, "tokens_per_second": 150596.93019013514, "data_wait_s": 7.747999916318804e-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:f9762b09b62efba49c5760b197cd594820f1a0953d49e57386c80c929e0ed9da
-size 676269
+oid sha256:e5b8bff9eadbb3a9c4366a1776d03bb90c27041832da9d7ae5a86faf01baa155
+size 678387