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": 76891.00562996072,
-      "required_pages_per_second_upper_estimate": 0.005319498181194081
+      "required_token_positions_per_second": 76275.23189235547,
+      "required_pages_per_second_upper_estimate": 0.00527689752523463
     },
     {
       "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": 9611.37570374509,
-      "required_pages_per_second_upper_estimate": 0.004024082298810068
+      "required_token_positions_per_second": 9534.403986544434,
+      "required_pages_per_second_upper_estimate": 0.003991855848170386
     },
     {
       "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": 19222.75140749018,
-      "required_pages_per_second_upper_estimate": 0.004003592180135482
+      "required_token_positions_per_second": 19068.807973088868,
+      "required_pages_per_second_upper_estimate": 0.003971529822511061
     },
     {
       "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": 28834.127111235273,
-      "required_pages_per_second_upper_estimate": 0.0017285618931557113
+      "required_token_positions_per_second": 28603.2119596333,
+      "required_pages_per_second_upper_estimate": 0.0017147188824041938
     },
     {
       "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": 28834.127111235273,
-      "required_pages_per_second_upper_estimate": 0.0022417265549769683
+      "required_token_positions_per_second": 28603.2119596333,
+      "required_pages_per_second_upper_estimate": 0.0022237739176283254
     },
     {
       "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": 28834.127111235273,
-      "required_pages_per_second_upper_estimate": 0.3698532229093428
+      "required_token_positions_per_second": 28603.2119596333,
+      "required_pages_per_second_upper_estimate": 0.36689129128196535
     }
   ],
   "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": 0.08977925437517342,
-    "producer_p95_seconds": 0.21518766700319247,
-    "gpu_update_seconds": 0.9588637760158484,
-    "producer_batches_per_second": 11.138430664852226,
-    "gpu_batches_per_second": 1.0429010095209517,
-    "producer_headroom": 10.680237686190921,
+    "producer_batch_seconds": 0.08202274746102489,
+    "producer_p95_seconds": 0.22649603500030935,
+    "gpu_update_seconds": 0.9666047309308704,
+    "producer_batches_per_second": 12.191739864300137,
+    "gpu_batches_per_second": 1.0345490436788665,
+    "producer_headroom": 11.784593431111,
     "sustainable": true,
-    "buffer_seconds": 60.40841788899845,
+    "buffer_seconds": 60.89609804864484,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 283401.03202731785,
+    "generation_bytes_per_second": 284166.3542818683,
     "upload_bytes_per_second": 5451365.219749806,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.811264234682575,
+    "estimated_full_shards_per_hour": 60.97548457471883,
     "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-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -189,3 +189,4 @@
 {"step": 543900, "loss": 0.5678874254226685, "lr": 0.0003, "grad_norm": 0.15976679832301713, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79609342577, "tokens_per_second": 148429.45422243042, "data_wait_s": 9.993000276153907e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 544000, "loss": 0.5968501567840576, "lr": 0.0003, "grad_norm": 0.1837794646857543, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79624080264, "tokens_per_second": 153831.34480019816, "data_wait_s": 9.872001101030037e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 544100, "loss": 0.5709521174430847, "lr": 0.0003, "grad_norm": 0.16342055432747546, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79638776422, "tokens_per_second": 153453.21711935665, "data_wait_s": 1.3431999832391739e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 544200, "loss": 0.6047398447990417, "lr": 0.0003, "grad_norm": 0.15733616336518852, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79653472941, "tokens_per_second": 153841.05317850888, "data_wait_s": 9.531999239698052e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a83b2a82196af432bb591e6bef76a334b8de6b39667582a5072739ea57021cf8
-size 407052
+oid sha256:4b443db417f03148b31fea71e555d327039b2ffbdb6dc3f31357d7ea0a1ee980
+size 409170