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": 77116.80661270273,
-      "required_pages_per_second_upper_estimate": 0.005335119617110637
+      "required_token_positions_per_second": 76702.70264228176,
+      "required_pages_per_second_upper_estimate": 0.005306470943583323
     },
     {
       "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": 9639.60082658784,
-      "required_pages_per_second_upper_estimate": 0.004035899568336739
+      "required_token_positions_per_second": 9587.83783028522,
+      "required_pages_per_second_upper_estimate": 0.004014227482719112
     },
     {
       "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": 19279.20165317568,
-      "required_pages_per_second_upper_estimate": 0.004015349277618683
+      "required_token_positions_per_second": 19175.67566057044,
+      "required_pages_per_second_upper_estimate": 0.003993787543523032
     },
     {
       "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": 28918.802479763523,
-      "required_pages_per_second_upper_estimate": 0.0017336380522071783
+      "required_token_positions_per_second": 28763.51349085566,
+      "required_pages_per_second_upper_estimate": 0.0017243287144347087
     },
     {
       "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": 28918.802479763523,
-      "required_pages_per_second_upper_estimate": 0.0022483096924324547
+      "required_token_positions_per_second": 28763.51349085566,
+      "required_pages_per_second_upper_estimate": 0.0022362366565889446
     },
     {
       "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": 28918.802479763523,
-      "required_pages_per_second_upper_estimate": 0.37093934761949593
+      "required_token_positions_per_second": 28763.51349085566,
+      "required_pages_per_second_upper_estimate": 0.36894746720611155
     }
   ],
   "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": 0.08553106301565094,
-    "producer_p95_seconds": 0.20938527499856718,
-    "gpu_update_seconds": 0.956056185913895,
-    "producer_batches_per_second": 11.691658734757155,
-    "gpu_batches_per_second": 1.0459636313571876,
-    "producer_headroom": 11.1778826569588,
+    "producer_batch_seconds": 0.3635166038357909,
+    "producer_p95_seconds": 0.24673293700107024,
+    "gpu_update_seconds": 0.9612177597423799,
+    "producer_batches_per_second": 2.750905981867403,
+    "gpu_batches_per_second": 1.0403469867931012,
+    "producer_headroom": 2.644219685152497,
     "sustainable": true,
-    "buffer_seconds": 60.23153971257538,
+    "buffer_seconds": 60.55671886376993,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 340442.04527818813,
+    "generation_bytes_per_second": 323291.7802778824,
     "upload_bytes_per_second": 5376747.698464667,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.05093783148988,
+    "estimated_full_shards_per_hour": 69.37089020016055,
     "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

@@ -133,3 +133,4 @@
 {"step": 410400, "loss": 0.5478192567825317, "lr": 0.0003, "grad_norm": 0.14434415254926053, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60071842138, "tokens_per_second": 153785.14796797503, "data_wait_s": 1.5442999938386492e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 410500, "loss": 0.5681799054145813, "lr": 0.0003, "grad_norm": 0.1623207670027057, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60086465650, "tokens_per_second": 150977.93994714442, "data_wait_s": 2.4000999474083073e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 410600, "loss": 0.6016080975532532, "lr": 0.0003, "grad_norm": 0.1727383479707727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60101196151, "tokens_per_second": 152161.6338070798, "data_wait_s": 1.414700091117993e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 410700, "loss": 0.6070379018783569, "lr": 0.0003, "grad_norm": 0.16094560335790287, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60115921087, "tokens_per_second": 152569.10342319703, "data_wait_s": 1.0637000741553493e-05, "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 sha256:72e866e5028e32a578ea3f99b9c1202fa1971a659949e1639a8ac177f17db35b
-size 288444
+oid sha256:738bb476ec6833e0c68e20d8f6d6660602f7fa342c4538658bd5edd44ce6e513
+size 290562