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": 77108.74430827609,
-      "required_pages_per_second_upper_estimate": 0.005334561848183279
+      "required_token_positions_per_second": 77005.85091025478,
+      "required_pages_per_second_upper_estimate": 0.005327443444162588
     },
     {
       "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": 9638.593038534511,
-      "required_pages_per_second_upper_estimate": 0.004035477628523817
+      "required_token_positions_per_second": 9625.731363781848,
+      "required_pages_per_second_upper_estimate": 0.004030092713887188
     },
     {
       "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": 19277.186077069022,
-      "required_pages_per_second_upper_estimate": 0.004014929486269982
+      "required_token_positions_per_second": 19251.462727563696,
+      "required_pages_per_second_upper_estimate": 0.004009571990938368
     },
     {
       "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": 28915.779115603535,
-      "required_pages_per_second_upper_estimate": 0.0017334568061421968
+      "required_token_positions_per_second": 28877.194091345544,
+      "required_pages_per_second_upper_estimate": 0.0017311436928538508
     },
     {
       "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": 28915.779115603535,
-      "required_pages_per_second_upper_estimate": 0.0022480746391673892
+      "required_token_positions_per_second": 28877.194091345544,
+      "required_pages_per_second_upper_estimate": 0.0022450748232489158
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28915.779115603535,
-      "required_pages_per_second_upper_estimate": 0.37090056715028713
+      "required_token_positions_per_second": 28877.194091345544,
+      "required_pages_per_second_upper_estimate": 0.37040563988847686
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.06676878492174865,
-    "producer_p95_seconds": 0.24878291700224509,
-    "gpu_update_seconds": 0.9561561488440263,
-    "producer_batches_per_second": 14.977058533744101,
-    "gpu_batches_per_second": 1.045854279354873,
-    "producer_headroom": 14.320406608636318,
+    "producer_batch_seconds": 0.11066954131234752,
+    "producer_p95_seconds": 0.2435802030013292,
+    "gpu_update_seconds": 0.9574337420922093,
+    "producer_batches_per_second": 9.035909864103042,
+    "gpu_batches_per_second": 1.0444586983270234,
+    "producer_headroom": 8.651284994396082,
     "sustainable": true,
-    "buffer_seconds": 60.23783737717366,
+    "buffer_seconds": 60.31832575180919,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 365477.74109533447,
+    "generation_bytes_per_second": 438679.55824335857,
     "upload_bytes_per_second": 4856509.547001987,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.42301535267855,
+    "estimated_full_shards_per_hour": 94.13042126155442,
     "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-v1/1790824801708454555-41033672/train.jsonl

@@ -204,3 +204,4 @@
 {"step": 417500, "loss": 0.5674142241477966, "lr": 0.0003, "grad_norm": 0.13949844038831657, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61111418059, "tokens_per_second": 154139.1731713443, "data_wait_s": 1.004899968393147e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 417600, "loss": 0.5719407796859741, "lr": 0.0003, "grad_norm": 0.16183065922302256, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61125960944, "tokens_per_second": 146780.3492795418, "data_wait_s": 1.6945999959716573e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 417700, "loss": 0.6011258363723755, "lr": 0.0003, "grad_norm": 0.19596748567520328, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61140454233, "tokens_per_second": 149981.84880119187, "data_wait_s": 1.3928998669143766e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 417800, "loss": 0.5787203311920166, "lr": 0.0003, "grad_norm": 0.1585995637576139, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61154935523, "tokens_per_second": 150605.2872722367, "data_wait_s": 9.880001016426831e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a1454510cd79e222e05eeb7dccf0567b8ab2585289886acc342ec7b8773a61d0
-size 439605
+oid sha256:22ea88f0899c30519e6905a426534bf10510c675aebf12ba665a70745b746860
+size 441723