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": 75618.08761996537,
-      "required_pages_per_second_upper_estimate": 0.006554384182607696
+      "required_token_positions_per_second": 74750.89067393649,
+      "required_pages_per_second_upper_estimate": 0.006479217749216479
     },
     {
       "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": 9452.260952495672,
-      "required_pages_per_second_upper_estimate": 0.003770648217845728
+      "required_token_positions_per_second": 9343.861334242061,
+      "required_pages_per_second_upper_estimate": 0.0037274059894056412
     },
     {
       "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": 18904.521904991343,
-      "required_pages_per_second_upper_estimate": 0.0043732284344976165
+      "required_token_positions_per_second": 18687.722668484123,
+      "required_pages_per_second_upper_estimate": 0.004323075746667917
     },
     {
       "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": 28356.782857487015,
-      "required_pages_per_second_upper_estimate": 0.0017267792994793629
+      "required_token_positions_per_second": 28031.584002726184,
+      "required_pages_per_second_upper_estimate": 0.0017069763954109581
     },
     {
       "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": 28356.782857487015,
-      "required_pages_per_second_upper_estimate": 0.0021829836409683057
+      "required_token_positions_per_second": 28031.584002726184,
+      "required_pages_per_second_upper_estimate": 0.0021579489329207713
     },
     {
       "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": 28356.782857487015,
-      "required_pages_per_second_upper_estimate": 0.3520963390427631
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28031.584002726184,
+      "required_pages_per_second_upper_estimate": 0.3480584576371877
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.06377971641407854,
-    "producer_p95_seconds": 0.14048611499993058,
-    "gpu_update_seconds": 0.9750048212080633,
-    "producer_batches_per_second": 15.678965919316367,
-    "gpu_batches_per_second": 1.0256359540468394,
-    "producer_headroom": 15.287067362890372,
+    "producer_batch_seconds": 0.36570967231250684,
+    "producer_p95_seconds": 0.14958367799999905,
+    "gpu_update_seconds": 0.986316006876783,
+    "producer_batches_per_second": 2.7344094939481893,
+    "gpu_batches_per_second": 1.013873842691196,
+    "producer_headroom": 2.696991853236943,
     "sustainable": true,
-    "buffer_seconds": 60.450298914899925,
+    "buffer_seconds": 60.16527641948376,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 62,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 358974.2303335235,
-    "upload_bytes_per_second": 4369690.186918407,
+    "generation_bytes_per_second": 314817.22246318957,
+    "upload_bytes_per_second": 4415852.391526113,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.02751333717612,
+    "estimated_full_shards_per_hour": 67.55244737073674,
     "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": [
@@ -220,6 +220,12 @@
       "bytes": 6008731,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.3192515309999635,
+      "bytes": 6183122,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -8,3 +8,4 @@
 {"step": 315800, "loss": 0.5816597454249859, "lr": 0.0003, "grad_norm": 0.15681656811833933, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46224346598, "tokens_per_second": 148903.31554262608, "data_wait_s": 1.0683000027711387e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 315900, "loss": 0.5567959807813168, "lr": 0.0003, "grad_norm": 0.15135116279330912, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46239028327, "tokens_per_second": 149536.34382585716, "data_wait_s": 1.0694999900806579e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 316000, "loss": 0.5752813629806042, "lr": 0.0003, "grad_norm": 0.1705854728676636, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46253660338, "tokens_per_second": 147117.6807879494, "data_wait_s": 7.873000186009449e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 316100, "loss": 0.5824047094210982, "lr": 0.0003, "grad_norm": 0.15162778451768252, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46268355495, "tokens_per_second": 129192.18210135636, "data_wait_s": 1.2212000001454726e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:09c6be592c1962a0f1b6e0ca684d0396518b545f898c6b786b33bf1bf36d876b
-size 21345
+oid sha256:67091f6c09a4c04766e42cd1597d892fb1cf7967c366eebd0543b82e49226589
+size 23463