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": 75644.40111509674,
-      "required_pages_per_second_upper_estimate": 0.006556664969674739
+      "required_token_positions_per_second": 75595.3099323377,
+      "required_pages_per_second_upper_estimate": 0.00655240987037365
     },
     {
       "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": 9455.550139387093,
-      "required_pages_per_second_upper_estimate": 0.0037719603236744427
+      "required_token_positions_per_second": 9449.413741542212,
+      "required_pages_per_second_upper_estimate": 0.0037695124228267957
     },
     {
       "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": 18911.100278774185,
-      "required_pages_per_second_upper_estimate": 0.004374750225496852
+      "required_token_positions_per_second": 18898.827483084424,
+      "required_pages_per_second_upper_estimate": 0.004371911130207853
     },
     {
       "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": 28366.650418161276,
-      "required_pages_per_second_upper_estimate": 0.0017273801821533357
+      "required_token_positions_per_second": 28348.241224626632,
+      "required_pages_per_second_upper_estimate": 0.0017262591588526497
     },
     {
       "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": 28366.650418161276,
-      "required_pages_per_second_upper_estimate": 0.0021837432731041666
+      "required_token_positions_per_second": 28348.241224626632,
+      "required_pages_per_second_upper_estimate": 0.0021823260824259606
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28366.650418161276,
-      "required_pages_per_second_upper_estimate": 0.3522188611217363
+      "required_token_positions_per_second": 28348.241224626632,
+      "required_pages_per_second_upper_estimate": 0.35199028054964343
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.11311328864837833,
-    "producer_p95_seconds": 0.3479473679981311,
-    "gpu_update_seconds": 0.9746656581736851,
-    "producer_batches_per_second": 8.84069424511721,
-    "gpu_batches_per_second": 1.025992853666134,
-    "producer_headroom": 8.616721075129474,
+    "producer_batch_seconds": 0.09540040973456598,
+    "producer_p95_seconds": 0.3536560149950674,
+    "gpu_update_seconds": 0.9752986007464082,
+    "producer_batches_per_second": 10.482135273656741,
+    "gpu_batches_per_second": 1.0253270118860907,
+    "producer_headroom": 10.223211865231988,
     "sustainable": true,
-    "buffer_seconds": 60.42927080676848,
+    "buffer_seconds": 60.468513246277304,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 335883.93137144716,
-    "upload_bytes_per_second": 5235517.812625838,
+    "generation_bytes_per_second": 334380.1806155741,
+    "upload_bytes_per_second": 5209264.115662761,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.07287269456445,
+    "estimated_full_shards_per_hour": 71.75020278788011,
     "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": 0.9317472719994839,
-      "bytes": 2642711,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.9964947309999843,
       "bytes": 5697842,
@@ -562,6 +556,12 @@
       "bytes": 10467154,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.0654726889988524,
+      "bytes": 1251369,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -380,3 +380,4 @@
 {"step": 275100, "loss": 0.5664983093738556, "lr": 0.0003, "grad_norm": 0.15217281702798632, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40266125057, "tokens_per_second": 137468.61032182942, "data_wait_s": 1.0296003893017769e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 275200, "loss": 0.551712142303586, "lr": 0.0003, "grad_norm": 0.14366926736023933, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40280693026, "tokens_per_second": 147906.70054979614, "data_wait_s": 1.1684998753480613e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 275300, "loss": 0.571684068068862, "lr": 0.0003, "grad_norm": 0.13297020996572717, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40295403551, "tokens_per_second": 150699.85841685484, "data_wait_s": 1.8714999896474183e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 275400, "loss": 0.5842844024300575, "lr": 0.0003, "grad_norm": 0.15522975680030643, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40310118451, "tokens_per_second": 150884.40286893057, "data_wait_s": 1.2607000826392323e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:34c4d03ffd36a7fe747aac8bddfac6a5dfe811bbc849c6321b52427d3b25371e
-size 815505
+oid sha256:c67555fd355921459881473c4405ba484c66ebc3c66772eac8bd9be790c7622f
+size 817623