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": 75707.57068446348,
-      "required_pages_per_second_upper_estimate": 0.006562140347845639
+      "required_token_positions_per_second": 75982.01804245898,
+      "required_pages_per_second_upper_estimate": 0.006585928749256219
     },
     {
       "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": 9463.446335557936,
-      "required_pages_per_second_upper_estimate": 0.0037751102343856454
+      "required_token_positions_per_second": 9497.752255307372,
+      "required_pages_per_second_upper_estimate": 0.003788795378692904
     },
     {
       "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": 18926.89267111587,
-      "required_pages_per_second_upper_estimate": 0.004378403517528489
+      "required_token_positions_per_second": 18995.504510614745,
+      "required_pages_per_second_upper_estimate": 0.00439427565906942
     },
     {
       "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": 28390.339006673807,
-      "required_pages_per_second_upper_estimate": 0.00172882269290933
+      "required_token_positions_per_second": 28493.256765922113,
+      "required_pages_per_second_upper_estimate": 0.0017350898444798
     },
     {
       "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": 28390.339006673807,
-      "required_pages_per_second_upper_estimate": 0.0021855668862220713
+      "required_token_positions_per_second": 28493.256765922113,
+      "required_pages_per_second_upper_estimate": 0.0021934897802235974
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28390.339006673807,
-      "required_pages_per_second_upper_estimate": 0.35251299411045617
+      "required_token_positions_per_second": 28493.256765922113,
+      "required_pages_per_second_upper_estimate": 0.3537908882367373
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.1626985590626191,
-    "producer_p95_seconds": 0.42259284100146033,
-    "gpu_update_seconds": 0.9738524078032565,
-    "producer_batches_per_second": 6.146335934143842,
-    "gpu_batches_per_second": 1.0268496457853662,
-    "producer_headroom": 5.9856240486336585,
+    "producer_batch_seconds": 0.10824300639103512,
+    "producer_p95_seconds": 0.41639111100084847,
+    "gpu_update_seconds": 0.9703348489480838,
+    "producer_batches_per_second": 9.238472150223108,
+    "gpu_batches_per_second": 1.0305720763137338,
+    "producer_headroom": 8.96441147839782,
     "sustainable": true,
-    "buffer_seconds": 60.378849283801905,
+    "buffer_seconds": 60.16076063478119,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 540097.3082140611,
+    "generation_bytes_per_second": 543027.0978601879,
     "upload_bytes_per_second": 5817395.304760269,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 115.89230952087759,
+    "estimated_full_shards_per_hour": 116.52097417692401,
     "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-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -420,3 +420,4 @@
 {"step": 157000, "loss": 0.6234585996717215, "lr": 0.0003, "grad_norm": 0.15615076115668247, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22977662403, "tokens_per_second": 148993.8608351398, "data_wait_s": 1.464000524720177e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 157100, "loss": 0.6028835140168667, "lr": 0.0003, "grad_norm": 0.13410704850467242, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22992238879, "tokens_per_second": 147704.04496869977, "data_wait_s": 9.82900382950902e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 157200, "loss": 0.6465106811374426, "lr": 0.0003, "grad_norm": 0.15757927262753454, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23006794010, "tokens_per_second": 148872.62405405362, "data_wait_s": 1.040300412569195e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 157300, "loss": 0.587610375136137, "lr": 0.0003, "grad_norm": 0.15463878370965745, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23021545116, "tokens_per_second": 149662.42141581018, "data_wait_s": 1.5237004845403135e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:380c2f377b4e37ff1a08c492ba57b9dc36ae9c13964f7677a8216441b684bef5
-size 900225
+oid sha256:646616b2628ecafdca2162d132a88183d5a2e85ace521cdde130f3c131e63a2a
+size 902343