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": 77207.00454451541,
-      "required_pages_per_second_upper_estimate": 0.005341359719321474
+      "required_token_positions_per_second": 75816.91873395538,
+      "required_pages_per_second_upper_estimate": 0.0052451903575034715
     },
     {
       "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": 9650.875568064426,
-      "required_pages_per_second_upper_estimate": 0.0040406200671496105
+      "required_token_positions_per_second": 9477.114841744422,
+      "required_pages_per_second_upper_estimate": 0.003967870079576005
     },
     {
       "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": 19301.751136128853,
-      "required_pages_per_second_upper_estimate": 0.004020045740247973
+      "required_token_positions_per_second": 18954.229683488844,
+      "required_pages_per_second_upper_estimate": 0.0039476661864200685
     },
     {
       "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": 28952.62670419328,
-      "required_pages_per_second_upper_estimate": 0.001735665762815142
+      "required_token_positions_per_second": 28431.344525233264,
+      "required_pages_per_second_upper_estimate": 0.001704415692138289
     },
     {
       "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": 28952.62670419328,
-      "required_pages_per_second_upper_estimate": 0.0022509393771048287
+      "required_token_positions_per_second": 28431.344525233264,
+      "required_pages_per_second_upper_estimate": 0.002210411980568674
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28952.62670419328,
-      "required_pages_per_second_upper_estimate": 0.3713732084528582
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28431.344525233264,
+      "required_pages_per_second_upper_estimate": 0.36468676037035525
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07716357572658694,
-    "producer_p95_seconds": 0.17510002500057453,
-    "gpu_update_seconds": 0.9549392627645654,
-    "producer_batches_per_second": 12.959482379915775,
-    "gpu_batches_per_second": 1.047187019104213,
-    "producer_headroom": 12.375518549687145,
+    "producer_batch_seconds": 0.3311217208672588,
+    "producer_p95_seconds": 0.1832798310006183,
+    "gpu_update_seconds": 0.9724478550587703,
+    "producer_batches_per_second": 3.020037457466837,
+    "gpu_batches_per_second": 1.0283327736267818,
+    "producer_headroom": 2.9368289477107683,
     "sustainable": true,
-    "buffer_seconds": 60.16117355416762,
+    "buffer_seconds": 60.29176701364376,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 315936.5058649097,
+    "generation_bytes_per_second": 320739.2771486744,
     "upload_bytes_per_second": 3780439.5388559783,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.79261953316181,
+    "estimated_full_shards_per_hour": 68.82318244786428,
     "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-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -67,3 +67,4 @@
 {"step": 931700, "loss": 0.5648506879806519, "lr": 3e-05, "grad_norm": 0.33845658496693376, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136374738469, "tokens_per_second": 150816.21011358508, "data_wait_s": 1.3628000488097314e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 931800, "loss": 0.5013023614883423, "lr": 3e-05, "grad_norm": 0.3371349907046047, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136389199628, "tokens_per_second": 151124.8256593921, "data_wait_s": 1.2116999641875736e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 931900, "loss": 0.5080162286758423, "lr": 3e-05, "grad_norm": 0.34191860752855485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136403720237, "tokens_per_second": 151348.02595877246, "data_wait_s": 1.0497999937797431e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 932000, "loss": 0.4959946572780609, "lr": 3e-05, "grad_norm": 0.33699241477416325, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136418450202, "tokens_per_second": 153685.8528782513, "data_wait_s": 6.844999916211236e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fb79ce9b761fbf5cb92fabeefc1aaa1c977ed55578b2276bedb34952c4c93008
-size 147090
+oid sha256:7b697b7a5b01673e4584c31ec17d258b264a9c85bf0b51cb72ed63a591fd52cb
+size 149208