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": 76216.92050433328,
-      "required_pages_per_second_upper_estimate": 0.006606289499296637
+      "required_token_positions_per_second": 75929.00970984956,
+      "required_pages_per_second_upper_estimate": 0.006581334121334026
     },
     {
       "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": 9527.11506304166,
-      "required_pages_per_second_upper_estimate": 0.0038005086417112095
+      "required_token_positions_per_second": 9491.126213731195,
+      "required_pages_per_second_upper_estimate": 0.003786152151640017
     },
     {
       "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": 19054.23012608332,
-      "required_pages_per_second_upper_estimate": 0.004407860796672543
+      "required_token_positions_per_second": 18982.25242746239,
+      "required_pages_per_second_upper_estimate": 0.00439121002286082
     },
     {
       "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": 28581.34518912498,
-      "required_pages_per_second_upper_estimate": 0.001740453967288616
+      "required_token_positions_per_second": 28473.378641193587,
+      "required_pages_per_second_upper_estimate": 0.001733879370976293
     },
     {
       "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": 28581.34518912498,
-      "required_pages_per_second_upper_estimate": 0.002200271070886116
+      "required_token_positions_per_second": 28473.378641193587,
+      "required_pages_per_second_upper_estimate": 0.0021919595071558245
     },
     {
       "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": 28581.34518912498,
-      "required_pages_per_second_upper_estimate": 0.3548846516399292
+      "required_token_positions_per_second": 28473.378641193587,
+      "required_pages_per_second_upper_estimate": 0.35354406845541286
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07477059708589451,
-    "producer_p95_seconds": 0.17951727800027584,
-    "gpu_update_seconds": 0.9673442525903185,
-    "producer_batches_per_second": 13.37424119873252,
-    "gpu_batches_per_second": 1.0337581448612911,
-    "producer_headroom": 12.937495356350555,
+    "producer_batch_seconds": 0.10290743771874844,
+    "producer_p95_seconds": 0.18207166000047437,
+    "gpu_update_seconds": 0.9710122689830887,
+    "producer_batches_per_second": 9.717470594623624,
+    "gpu_batches_per_second": 1.0298531047885413,
+    "producer_headroom": 9.43578317086193,
     "sustainable": true,
-    "buffer_seconds": 60.94268791319007,
+    "buffer_seconds": 60.2027606769515,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 369886.6426906674,
+    "generation_bytes_per_second": 390672.94198554487,
     "upload_bytes_per_second": 6249709.309424514,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.36906300106065,
+    "estimated_full_shards_per_hour": 83.8293189494587,
     "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

@@ -59,3 +59,4 @@
 {"step": 45900, "loss": 0.6565524749457836, "lr": 0.0003, "grad_norm": 0.13112385998803777, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6716776605, "tokens_per_second": 150495.40776296184, "data_wait_s": 1.0531999578233808e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 46000, "loss": 0.6058008559048176, "lr": 0.0003, "grad_norm": 0.1322720212095008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6731396222, "tokens_per_second": 149629.9861653259, "data_wait_s": 1.0580999514786527e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 46100, "loss": 0.6462961342185736, "lr": 0.0003, "grad_norm": 0.12934043426875, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6745887842, "tokens_per_second": 148537.8203293296, "data_wait_s": 1.0992000170517713e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 46200, "loss": 0.6512833032757044, "lr": 0.0003, "grad_norm": 0.14297691303761664, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6760451345, "tokens_per_second": 148569.01473379356, "data_wait_s": 1.332300053036306e-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:aa9e4440416f2aafaa8369ed9b7fd52fe8c112c2cf64a7f2aeef931a3c697e19
-size 130146
+oid sha256:9a462f5a7e3ca2db2623c77ef267af45bdd7fb432e1b01f8a3bee910fca6e55d
+size 132264