Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74965.3973539503,
-      "required_pages_per_second_upper_estimate": 0.006497810644577866
+      "required_token_positions_per_second": 74477.16912629639,
+      "required_pages_per_second_upper_estimate": 0.006455492259207954
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9370.674669243788,
-      "required_pages_per_second_upper_estimate": 0.003738102229632914
+      "required_token_positions_per_second": 9309.646140787048,
+      "required_pages_per_second_upper_estimate": 0.0037137570371737066
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18741.349338487576,
-      "required_pages_per_second_upper_estimate": 0.004335481332976058
+      "required_token_positions_per_second": 18619.292281574097,
+      "required_pages_per_second_upper_estimate": 0.004307245580989959
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28112.02400773137,
-      "required_pages_per_second_upper_estimate": 0.0017118747696796843
+      "required_token_positions_per_second": 27928.938422361145,
+      "required_pages_per_second_upper_estimate": 0.001700725818106467
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28112.02400773137,
-      "required_pages_per_second_upper_estimate": 0.002164141427178256
+      "required_token_positions_per_second": 27928.938422361145,
+      "required_pages_per_second_upper_estimate": 0.002150046992003118
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28112.02400773137,
-      "required_pages_per_second_upper_estimate": 0.3490572532839734
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27928.938422361145,
+      "required_pages_per_second_upper_estimate": 0.34678394306171256
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.5699123814143832,
-    "producer_p95_seconds": 0.45538208500511246,
-    "gpu_update_seconds": 0.9834937531497643,
-    "producer_batches_per_second": 1.7546556849988844,
-    "gpu_batches_per_second": 1.0167832757425985,
-    "producer_headroom": 1.7256929051251235,
+    "producer_batch_seconds": 0.5124305633830204,
+    "producer_p95_seconds": 0.44285217400465626,
+    "gpu_update_seconds": 0.989940955932066,
+    "producer_batches_per_second": 1.9514839111041506,
+    "gpu_batches_per_second": 1.010161256595817,
+    "producer_headroom": 1.9318538484444898,
     "sustainable": true,
-    "buffer_seconds": 60.97661269528538,
+    "buffer_seconds": 60.38639831185603,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 261433.70408706358,
+    "generation_bytes_per_second": 245545.11968123144,
     "upload_bytes_per_second": 6253193.739659444,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.097587031926444,
+    "estimated_full_shards_per_hour": 52.68826668575008,
     "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-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -416,3 +416,4 @@
 {"step": 81600, "loss": 0.6269095726311207, "lr": 0.0003, "grad_norm": 0.13646336345421883, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11941723632, "tokens_per_second": 148375.01129323352, "data_wait_s": 8.81299638422206e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 81700, "loss": 0.7072742599993944, "lr": 0.0003, "grad_norm": 0.16792434377732804, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11956176838, "tokens_per_second": 145960.47523637753, "data_wait_s": 1.1839998478535563e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 81800, "loss": 0.6340364888310432, "lr": 0.0003, "grad_norm": 0.14381530252142383, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11970645618, "tokens_per_second": 129867.3593586112, "data_wait_s": 1.4462995750363916e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 81900, "loss": 0.603107076138258, "lr": 0.0003, "grad_norm": 0.12883536764548545, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11985304470, "tokens_per_second": 148491.37312651347, "data_wait_s": 1.3864999345969409e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0931769004599d9e2ef5449a7d92502afc604802b2024e3a6eab9cf6f2451d98
-size 891753
+oid sha256:c8c51aaa31e783404e2cf3f0bfe6da17cb3c791ce9dde228fe73e0abe59a0c3d
+size 893871