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": 76073.54609128118,
-      "required_pages_per_second_upper_estimate": 0.005262944433540777
+      "required_token_positions_per_second": 76760.87516472297,
+      "required_pages_per_second_upper_estimate": 0.005310495453664655
     },
     {
       "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": 9509.193261410148,
-      "required_pages_per_second_upper_estimate": 0.003981300644016468
+      "required_token_positions_per_second": 9595.10939559037,
+      "required_pages_per_second_upper_estimate": 0.004017271935264828
     },
     {
       "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": 19018.386522820296,
-      "required_pages_per_second_upper_estimate": 0.00396102836412318
+      "required_token_positions_per_second": 19190.21879118074,
+      "required_pages_per_second_upper_estimate": 0.003996816494101012
     },
     {
       "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": 28527.579784230442,
-      "required_pages_per_second_upper_estimate": 0.0017101848489724377
+      "required_token_positions_per_second": 28785.328186771112,
+      "required_pages_per_second_upper_estimate": 0.0017256364721457281
     },
     {
       "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": 28527.579784230442,
-      "required_pages_per_second_upper_estimate": 0.002217893848661535
+      "required_token_positions_per_second": 28785.328186771112,
+      "required_pages_per_second_upper_estimate": 0.0022379326532436642
     },
     {
       "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": 28527.579784230442,
-      "required_pages_per_second_upper_estimate": 0.36592116294340044
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28785.328186771112,
+      "required_pages_per_second_upper_estimate": 0.36922728270251937
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 0.09522889598414963,
+    "producer_batch_seconds": 0.08881600480449947,
     "producer_p95_seconds": 0.2770388940007251,
-    "gpu_update_seconds": 0.9691673885102353,
-    "producer_batches_per_second": 10.501014315722456,
-    "gpu_batches_per_second": 1.0318135049273163,
-    "producer_headroom": 10.17724062107733,
+    "gpu_update_seconds": 0.9604893097139051,
+    "producer_batches_per_second": 11.259231961640088,
+    "gpu_batches_per_second": 1.0411360021256912,
+    "producer_headroom": 10.814371934744425,
     "sustainable": true,
-    "buffer_seconds": 60.08837808763459,
+    "buffer_seconds": 60.51082651197602,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 451549.1141793604,
-    "upload_bytes_per_second": 4950490.843476476,
+    "generation_bytes_per_second": 462105.63759715285,
+    "upload_bytes_per_second": 4923098.401492882,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.89192837749108,
+    "estimated_full_shards_per_hour": 99.15711255966129,
     "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": [
-    {
-      "seconds": 1.473726465999789,
-      "bytes": 9735362,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.1486626710011478,
       "bytes": 2585051,
@@ -562,6 +556,12 @@
       "bytes": 6517566,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.2599434170006134,
+      "bytes": 6272622,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -234,3 +234,4 @@
 {"step": 908400, "loss": 0.5087893605232239, "lr": 3e-05, "grad_norm": 0.3203193908856255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132964035956, "tokens_per_second": 152927.8710040776, "data_wait_s": 1.669399716774933e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 908500, "loss": 0.526980996131897, "lr": 3e-05, "grad_norm": 0.3245610635813911, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132978672808, "tokens_per_second": 151831.38066959972, "data_wait_s": 1.5254001482389867e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 908600, "loss": 0.5326830744743347, "lr": 3e-05, "grad_norm": 0.32527918325677957, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132993122077, "tokens_per_second": 150627.49092308353, "data_wait_s": 1.9208000594517216e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 908700, "loss": 0.5381379723548889, "lr": 3e-05, "grad_norm": 0.32745526137104025, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133007719042, "tokens_per_second": 151064.54576301624, "data_wait_s": 1.468199843657203e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bffafe52993495da51fc083f5316d7424e28b2e56bf2fe4ad3ca5e8925b1f1c4
-size 503145
+oid sha256:e47cbf521dea6330ec27ebd4be34d3ebab0080d402e840ccce7eb0751c733b21
+size 505263