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": 75931.71762100208,
-      "required_pages_per_second_upper_estimate": 0.0052531324108794795
+      "required_token_positions_per_second": 77373.46996714444,
+      "required_pages_per_second_upper_estimate": 0.005352876183511946
     },
     {
       "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": 9491.46470262526,
-      "required_pages_per_second_upper_estimate": 0.003973878066667641
+      "required_token_positions_per_second": 9671.683745893055,
+      "required_pages_per_second_upper_estimate": 0.004049332016682293
     },
     {
       "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": 18982.92940525052,
-      "required_pages_per_second_upper_estimate": 0.003953643581600466
+      "required_token_positions_per_second": 19343.36749178611,
+      "required_pages_per_second_upper_estimate": 0.004028713329634278
     },
     {
       "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": 28474.39410787578,
-      "required_pages_per_second_upper_estimate": 0.0017069964488848014
+      "required_token_positions_per_second": 29015.051237679167,
+      "required_pages_per_second_upper_estimate": 0.0017394080182808773
     },
     {
       "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": 28474.39410787578,
-      "required_pages_per_second_upper_estimate": 0.002213758896264027
+      "required_token_positions_per_second": 29015.051237679167,
+      "required_pages_per_second_upper_estimate": 0.0022557926100068514
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28474.39410787578,
-      "required_pages_per_second_upper_estimate": 0.3652389541934529
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 29015.051237679167,
+      "required_pages_per_second_upper_estimate": 0.3721739233421732
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10024818024996307,
-    "producer_p95_seconds": 0.16779082099947118,
-    "gpu_update_seconds": 0.9709776403057087,
-    "producer_batches_per_second": 9.975243415955855,
-    "gpu_batches_per_second": 1.0298898331841646,
-    "producer_headroom": 9.685738313499874,
+    "producer_batch_seconds": 0.06825911769530535,
+    "producer_p95_seconds": 0.16989970800023002,
+    "gpu_update_seconds": 0.9528847553471178,
+    "producer_batches_per_second": 14.650057512665109,
+    "gpu_batches_per_second": 1.0494448508998542,
+    "producer_headroom": 13.959816468777097,
     "sustainable": true,
-    "buffer_seconds": 60.20061369895394,
+    "buffer_seconds": 60.03173958686842,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 311197.660340641,
-    "upload_bytes_per_second": 5888193.355163731,
+    "generation_bytes_per_second": 333072.82358383347,
+    "upload_bytes_per_second": 5829867.330134732,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.77577359833167,
+    "estimated_full_shards_per_hour": 71.46967440258267,
     "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": [
@@ -322,6 +322,12 @@
       "bytes": 2763911,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.3125206449994948,
+      "bytes": 6101597,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791033297667653750-ba4975eb/train.jsonl

@@ -40,3 +40,4 @@
 {"step": 604000, "loss": 0.5574530363082886, "lr": 0.0003, "grad_norm": 0.17135670402991998, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88407450940, "tokens_per_second": 150693.0444466796, "data_wait_s": 1.5667999832658097e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 604100, "loss": 0.557549774646759, "lr": 0.0003, "grad_norm": 0.1732923168330613, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88422108370, "tokens_per_second": 152041.5890907132, "data_wait_s": 7.935999747132882e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 604200, "loss": 0.58061683177948, "lr": 0.0003, "grad_norm": 0.1700130865377676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88436879538, "tokens_per_second": 154242.60399963276, "data_wait_s": 8.957999853009824e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 604300, "loss": 0.5678399801254272, "lr": 0.0003, "grad_norm": 0.20809004478359064, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88451685598, "tokens_per_second": 146077.92891373046, "data_wait_s": 1.169400002254406e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791033297.83ac8725-d818-41af-bbb1-78dcd88792cf-7zwhs.737.0.1791033297667653750-ba4975eb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b9dce4e8dfbb71f5b4390e631d1c1fdc740152a45333778e750ac1ce6bbd7972
-size 89121
+oid sha256:54f871ac2b4582dae9d1215464bcfe8974917c3f3fb1c47255aafb6dc5cb3299
+size 91239