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": 76030.4852263245,
-      "required_pages_per_second_upper_estimate": 0.006590129761914728
+      "required_token_positions_per_second": 76295.37193780951,
+      "required_pages_per_second_upper_estimate": 0.0066130894707170165
     },
     {
       "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": 9503.810653290562,
-      "required_pages_per_second_upper_estimate": 0.0037912121642295206
+      "required_token_positions_per_second": 9536.921492226189,
+      "required_pages_per_second_upper_estimate": 0.0038044205729321004
     },
     {
       "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": 19007.621306581124,
-      "required_pages_per_second_upper_estimate": 0.004397078666567917
+      "required_token_positions_per_second": 19073.842984452378,
+      "required_pages_per_second_upper_estimate": 0.004412397886281706
     },
     {
       "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": 28511.431959871687,
-      "required_pages_per_second_upper_estimate": 0.001736196618433456
+      "required_token_positions_per_second": 28610.764476678567,
+      "required_pages_per_second_upper_estimate": 0.0017422454475495566
     },
     {
       "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": 28511.431959871687,
-      "required_pages_per_second_upper_estimate": 0.0021948889569659924
+      "required_token_positions_per_second": 28610.764476678567,
+      "required_pages_per_second_upper_estimate": 0.0022025358490797915
     },
     {
       "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": 28511.431959871687,
-      "required_pages_per_second_upper_estimate": 0.35401656331712983
+      "required_token_positions_per_second": 28610.764476678567,
+      "required_pages_per_second_upper_estimate": 0.3552499407313231
     }
   ],
   "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": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.2402159922651208,
-    "producer_p95_seconds": 0.3625543289963389,
-    "gpu_update_seconds": 0.9697162892033301,
-    "producer_batches_per_second": 4.1629201726766105,
-    "gpu_batches_per_second": 1.0312294545671183,
-    "producer_headroom": 4.036851502097648,
+    "producer_batch_seconds": 0.11577524164027864,
+    "producer_p95_seconds": 0.352708274993347,
+    "gpu_update_seconds": 0.9663495717682293,
+    "producer_batches_per_second": 8.63742528913968,
+    "gpu_batches_per_second": 1.0348222105280152,
+    "producer_headroom": 8.346772229340203,
     "sustainable": true,
-    "buffer_seconds": 60.122409930606466,
+    "buffer_seconds": 60.88002302139845,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 399226.5307569339,
+    "generation_bytes_per_second": 399339.8377125002,
     "upload_bytes_per_second": 5697066.613027273,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.66471998244297,
+    "estimated_full_shards_per_hour": 85.68903301745657,
     "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-v1/1790703713325746110-adad6d18/train.jsonl

@@ -295,3 +295,4 @@
 {"step": 309700, "loss": 0.6228529755026102, "lr": 0.0003, "grad_norm": 0.15280441359725982, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45331890859, "tokens_per_second": 149748.29777745705, "data_wait_s": 1.4021003153175116e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 309800, "loss": 0.5477604027837515, "lr": 0.0003, "grad_norm": 0.15359239092624632, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45346368767, "tokens_per_second": 148047.64556304982, "data_wait_s": 1.1169002391397953e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 309900, "loss": 0.6197275631129742, "lr": 0.0003, "grad_norm": 0.16832917274897194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45360906797, "tokens_per_second": 149268.5298529645, "data_wait_s": 1.0035000741481781e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 310000, "loss": 0.5801891293376684, "lr": 0.0003, "grad_norm": 0.14141978433655597, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45375546310, "tokens_per_second": 150883.49656436348, "data_wait_s": 1.284800237044692e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d0b37df3ead78d03db464c656d7fdcf9efca77b7b6bcb8c69c361ae106409e67
-size 633126
+oid sha256:63cd6dcdb41cdb3d8142c03ef590804c6735ef19be8b3d7326cf214e8781719c
+size 635244