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": 75871.41155015722,
-      "required_pages_per_second_upper_estimate": 0.006576341658833094
+      "required_token_positions_per_second": 74602.20012608299,
+      "required_pages_per_second_upper_estimate": 0.006466329629381285
     },
     {
       "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": 9483.926443769653,
-      "required_pages_per_second_upper_estimate": 0.0037832800557562043
+      "required_token_positions_per_second": 9325.275015760373,
+      "required_pages_per_second_upper_estimate": 0.0037199916290730707
     },
     {
       "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": 18967.852887539306,
-      "required_pages_per_second_upper_estimate": 0.004387878942722862
+      "required_token_positions_per_second": 18650.550031520746,
+      "required_pages_per_second_upper_estimate": 0.004314476511322505
     },
     {
       "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": 28451.779331308957,
-      "required_pages_per_second_upper_estimate": 0.0017325640863271401
+      "required_token_positions_per_second": 27975.825047281116,
+      "required_pages_per_second_upper_estimate": 0.0017035809675689838
     },
     {
       "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": 28451.779331308957,
-      "required_pages_per_second_upper_estimate": 0.002190296732490189
+      "required_token_positions_per_second": 27975.825047281116,
+      "required_pages_per_second_upper_estimate": 0.0021536564541799422
     },
     {
       "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": 28451.779331308957,
-      "required_pages_per_second_upper_estimate": 0.3532758773148858
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27975.825047281116,
+      "required_pages_per_second_upper_estimate": 0.3473661180237793
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 0.43970151343717134,
-    "producer_p95_seconds": 0.3460549719966366,
-    "gpu_update_seconds": 0.9717494177798413,
-    "producer_batches_per_second": 2.2742700887766887,
-    "gpu_batches_per_second": 1.0290718797493112,
-    "producer_headroom": 2.210020634642855,
+    "producer_batch_seconds": 0.2140537983123636,
+    "producer_p95_seconds": 0.3658770080000977,
+    "gpu_update_seconds": 0.9882818452457767,
+    "producer_batches_per_second": 4.671722753271231,
+    "gpu_batches_per_second": 1.0118570980642765,
+    "producer_headroom": 4.616978783079572,
     "sustainable": true,
-    "buffer_seconds": 60.24846390235016,
+    "buffer_seconds": 60.285192559992375,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 272084.8481834995,
-    "upload_bytes_per_second": 4870957.399781466,
+    "generation_bytes_per_second": 268198.67309396865,
+    "upload_bytes_per_second": 4767939.973542438,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.383074609076864,
+    "estimated_full_shards_per_hour": 57.54919190038962,
     "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": 2.588106572999095,
-      "bytes": 12282184,
-      "pages": 39,
-      "shards": 4
-    },
     {
       "seconds": 2.1326607650007645,
       "bytes": 1150711,
@@ -562,6 +556,12 @@
       "bytes": 560796,
       "pages": 18,
       "shards": 1
+    },
+    {
+      "seconds": 4.375384323997423,
+      "bytes": 11988381,
+      "pages": 43,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -326,3 +326,4 @@
 {"step": 269700, "loss": 0.6271895542740822, "lr": 0.0003, "grad_norm": 0.20498917692522642, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39476135063, "tokens_per_second": 147992.2800459298, "data_wait_s": 1.1707998055499047e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 269800, "loss": 0.6319512575864792, "lr": 0.0003, "grad_norm": 0.149184925699109, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39490644177, "tokens_per_second": 147836.01482647393, "data_wait_s": 1.0606003343127668e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 269900, "loss": 0.6347472965717316, "lr": 0.0003, "grad_norm": 0.18366539029727452, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39505126761, "tokens_per_second": 132285.73974714748, "data_wait_s": 1.4253004337660968e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 270000, "loss": 0.6172356083989143, "lr": 0.0003, "grad_norm": 0.14430560851862967, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39519825793, "tokens_per_second": 149868.97792367495, "data_wait_s": 1.539799995953217e-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 sha256:84f8bb631d8de546e4276264102a4f5094039e1534b224680bf43fcc333f2e69
-size 699567
+oid sha256:85d81dd16847c9dba6b33d92f88acc8ebb9bc6ab1ad809c52f32a67e4b0dba9c
+size 701685