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": 75123.94275356123,
-      "required_pages_per_second_upper_estimate": 0.0065115529579864995
+      "required_token_positions_per_second": 74600.37100032733,
+      "required_pages_per_second_upper_estimate": 0.006466171085396661
     },
     {
       "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": 9390.492844195154,
-      "required_pages_per_second_upper_estimate": 0.0037460079959291343
+      "required_token_positions_per_second": 9325.046375040916,
+      "required_pages_per_second_upper_estimate": 0.0037199004208715958
     },
     {
       "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": 18780.98568839031,
-      "required_pages_per_second_upper_estimate": 0.004344650504950122
+      "required_token_positions_per_second": 18650.092750081832,
+      "required_pages_per_second_upper_estimate": 0.004314370727309492
     },
     {
       "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": 28171.478532585465,
-      "required_pages_per_second_upper_estimate": 0.0017154952383095145
+      "required_token_positions_per_second": 27975.139125122747,
+      "required_pages_per_second_upper_estimate": 0.0017035391985082944
     },
     {
       "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": 28171.478532585465,
-      "required_pages_per_second_upper_estimate": 0.0021687184010821827
+      "required_token_positions_per_second": 27975.139125122747,
+      "required_pages_per_second_upper_estimate": 0.002153603649993437
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28171.478532585465,
-      "required_pages_per_second_upper_estimate": 0.34979547950116674
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27975.139125122747,
+      "required_pages_per_second_upper_estimate": 0.3473576011662062
     }
   ],
   "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": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.455095377023639,
-    "producer_p95_seconds": 0.3449882000022626,
-    "gpu_update_seconds": 0.9814181377814457,
-    "producer_batches_per_second": 2.197341591426575,
-    "gpu_batches_per_second": 1.0189336853510367,
-    "producer_headroom": 2.1565108927275873,
+    "producer_batch_seconds": 0.25633829831252797,
+    "producer_p95_seconds": 0.3414044359997206,
+    "gpu_update_seconds": 0.988306076918525,
+    "producer_batches_per_second": 3.9010947899045454,
+    "gpu_batches_per_second": 1.0118322889584326,
+    "producer_headroom": 3.855475687497859,
     "sustainable": true,
-    "buffer_seconds": 60.84792454244963,
+    "buffer_seconds": 60.28667069203003,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 354978.0593804336,
+    "generation_bytes_per_second": 346120.8382687419,
     "upload_bytes_per_second": 4115146.977753103,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.17002807674174,
+    "estimated_full_shards_per_hour": 74.26947461172765,
     "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-v1/1790485060735632541-7579de36/train.jsonl

@@ -215,3 +215,4 @@
 {"step": 106500, "loss": 0.5803993083536625, "lr": 0.0003, "grad_norm": 0.12976601303436522, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15584311756, "tokens_per_second": 147872.0377700266, "data_wait_s": 1.5667999832658097e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 106600, "loss": 0.5847213715314865, "lr": 0.0003, "grad_norm": 0.13213030274642967, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15599028513, "tokens_per_second": 148926.07728392465, "data_wait_s": 1.3206001312937587e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 106700, "loss": 0.6291310619562864, "lr": 0.0003, "grad_norm": 0.139231069345712, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15613778785, "tokens_per_second": 148545.86893485233, "data_wait_s": 1.749500006553717e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 106800, "loss": 0.653808118775487, "lr": 0.0003, "grad_norm": 0.12560414814596255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15628534506, "tokens_per_second": 149153.49299840332, "data_wait_s": 1.3376000424614176e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6bad3ed2ffa7f5fab1289191e392f656e184d178583e8211dc5f6093f17a9db8
-size 462903
+oid sha256:c42247d68d25eee5e92b1d6bc0ea4eb2e41405888f02a7f02e58130e320f8f35
+size 465021