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": 74555.70822138175,
-      "required_pages_per_second_upper_estimate": 0.0064622998289144366
+      "required_token_positions_per_second": 75883.21674722266,
+      "required_pages_per_second_upper_estimate": 0.0065773649033947276
     },
     {
       "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": 9319.463527672719,
-      "required_pages_per_second_upper_estimate": 0.0037176733395854152
+      "required_token_positions_per_second": 9485.402093402832,
+      "required_pages_per_second_upper_estimate": 0.0037838687144578077
     },
     {
       "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": 18638.927055345437,
-      "required_pages_per_second_upper_estimate": 0.004311787740341742
+      "required_token_positions_per_second": 18970.804186805664,
+      "required_pages_per_second_upper_estimate": 0.004388561673867041
     },
     {
       "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": 27958.390583018154,
-      "required_pages_per_second_upper_estimate": 0.001702519300166934
+      "required_token_positions_per_second": 28456.2062802085,
+      "required_pages_per_second_upper_estimate": 0.0017328336642887163
     },
     {
       "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": 27958.390583018154,
-      "required_pages_per_second_upper_estimate": 0.0021523143008592934
+      "required_token_positions_per_second": 28456.2062802085,
+      "required_pages_per_second_upper_estimate": 0.0021906375312710545
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27958.390583018154,
-      "required_pages_per_second_upper_estimate": 0.3471496403270317
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28456.2062802085,
+      "required_pages_per_second_upper_estimate": 0.35333084520417324
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08360177253891266,
-    "producer_p95_seconds": 0.33020705899980385,
-    "gpu_update_seconds": 0.9888981240856302,
-    "producer_batches_per_second": 11.961468873576186,
-    "gpu_batches_per_second": 1.0112265112492098,
-    "producer_headroom": 11.828674130388144,
+    "producer_batch_seconds": 0.1934856878984874,
+    "producer_p95_seconds": 0.3408465909997176,
+    "gpu_update_seconds": 0.9715982421461918,
+    "producer_batches_per_second": 5.16834092930249,
+    "gpu_batches_per_second": 1.029231997982078,
+    "producer_headroom": 5.021550961722515,
     "sustainable": true,
-    "buffer_seconds": 60.32278556922344,
+    "buffer_seconds": 60.23909101306389,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 417809.7682826356,
+    "generation_bytes_per_second": 437727.9423537882,
     "upload_bytes_per_second": 5511460.304690656,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.65225015982915,
+    "estimated_full_shards_per_hour": 93.92622664413675,
     "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-v1/1790563279934888633-82973089/train.jsonl

@@ -242,3 +242,4 @@
 {"step": 179200, "loss": 0.6327262502163649, "lr": 0.0003, "grad_norm": 0.15070615208766142, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26227355581, "tokens_per_second": 149891.60798435472, "data_wait_s": 1.9053000869462267e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 179300, "loss": 0.6335432901978493, "lr": 0.0003, "grad_norm": 0.15791971190942056, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26241899685, "tokens_per_second": 147946.1481803719, "data_wait_s": 1.2898999557364732e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 179400, "loss": 0.6733214948326349, "lr": 0.0003, "grad_norm": 0.17627854405043109, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26256400691, "tokens_per_second": 148238.68769730596, "data_wait_s": 1.0067997209262103e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 179500, "loss": 0.5970980357378721, "lr": 0.0003, "grad_norm": 0.14066056975815575, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26271078539, "tokens_per_second": 148615.68477450014, "data_wait_s": 1.1324998922646046e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d385c8995a462629087db9bb66ac9f1ad25309a2993134e280e4f3841a248f71
-size 520089
+oid sha256:d28921edf7591b5052233cd25570fb7b7231db11acb627dc67b4f0d0a2fe8744
+size 522207