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": 76363.09675330724,
-      "required_pages_per_second_upper_estimate": 0.0052829762201898394
+      "required_token_positions_per_second": 76909.75008111288,
+      "required_pages_per_second_upper_estimate": 0.005320794965817907
     },
     {
       "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": 9545.387094163405,
-      "required_pages_per_second_upper_estimate": 0.0039964542459770816
+      "required_token_positions_per_second": 9613.71876013911,
+      "required_pages_per_second_upper_estimate": 0.004025063287593663
     },
     {
       "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": 19090.77418832681,
-      "required_pages_per_second_upper_estimate": 0.003976104805857073
+      "required_token_positions_per_second": 19227.43752027822,
+      "required_pages_per_second_upper_estimate": 0.004004568173848126
     },
     {
       "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": 28636.161282490215,
-      "required_pages_per_second_upper_estimate": 0.0017166941439987632
+      "required_token_positions_per_second": 28841.15628041733,
+      "required_pages_per_second_upper_estimate": 0.0017289832811152555
     },
     {
       "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": 28636.161282490215,
-      "required_pages_per_second_upper_estimate": 0.002226335582551696
+      "required_token_positions_per_second": 28841.15628041733,
+      "required_pages_per_second_upper_estimate": 0.0022422730419628255
     },
     {
       "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": 28636.161282490215,
-      "required_pages_per_second_upper_estimate": 0.36731392981734734
+      "required_token_positions_per_second": 28841.15628041733,
+      "required_pages_per_second_upper_estimate": 0.36994338554427636
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09339843210176468,
-    "producer_p95_seconds": 0.3408977729995968,
-    "gpu_update_seconds": 0.9654925367704772,
-    "producer_batches_per_second": 10.70681785011577,
-    "gpu_batches_per_second": 1.0357407871271056,
-    "producer_headroom": 10.337352726847703,
+    "producer_batch_seconds": 0.14674598210967815,
+    "producer_p95_seconds": 0.35029233200475574,
+    "gpu_update_seconds": 0.9586300816508018,
+    "producer_batches_per_second": 6.814496626235385,
+    "gpu_batches_per_second": 1.0431552474109278,
+    "producer_headroom": 6.532581457217141,
     "sustainable": true,
-    "buffer_seconds": 60.826029816540064,
+    "buffer_seconds": 60.39369514400051,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 398851.3690096873,
-    "upload_bytes_per_second": 4133622.907107098,
+    "generation_bytes_per_second": 414634.795985047,
+    "upload_bytes_per_second": 4102565.7014059834,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.58421900480236,
+    "estimated_full_shards_per_hour": 88.97097501433905,
     "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": [
-    {
-      "seconds": 1.793492888999026,
-      "bytes": 11753421,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.5141362069989555,
       "bytes": 6622032,
@@ -562,6 +556,12 @@
       "bytes": 5903925,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.512940188003995,
+      "bytes": 7372289,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -387,3 +387,4 @@
 {"step": 838700, "loss": 0.5397185683250427, "lr": 3e-05, "grad_norm": 0.28648062397296886, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122761678745, "tokens_per_second": 152947.86210918566, "data_wait_s": 9.016002877615392e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 838800, "loss": 0.5419321060180664, "lr": 3e-05, "grad_norm": 0.3079537050666023, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122776433900, "tokens_per_second": 152673.76832441418, "data_wait_s": 1.9810999219771475e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 838900, "loss": 0.553163468837738, "lr": 3e-05, "grad_norm": 0.3054196818727262, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122791213470, "tokens_per_second": 153108.42544575344, "data_wait_s": 1.0652001947164536e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 839000, "loss": 0.5086778998374939, "lr": 3e-05, "grad_norm": 0.281391932025953, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122805929564, "tokens_per_second": 151756.35776742973, "data_wait_s": 8.819995855446905e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2b597d6afb32b858cd2454a7616d4db7d88d68b52bce673a69cbf38f827345c6
-size 829548
+oid sha256:83b79c20ea6b84a3d88cd4818d873279184ca1ac3516ca40672ae72b6814a1a5
+size 831666