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": 74622.41249162675,
-      "required_pages_per_second_upper_estimate": 0.0064680815859988455
+      "required_token_positions_per_second": 74836.82821126148,
+      "required_pages_per_second_upper_estimate": 0.0064866665971451114
     },
     {
       "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": 9327.801561453343,
-      "required_pages_per_second_upper_estimate": 0.0037209995059252206
+      "required_token_positions_per_second": 9354.603526407685,
+      "required_pages_per_second_upper_estimate": 0.003731691210470594
     },
     {
       "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": 18655.603122906687,
-      "required_pages_per_second_upper_estimate": 0.004315645455083272
+      "required_token_positions_per_second": 18709.20705281537,
+      "required_pages_per_second_upper_estimate": 0.0043280457808707005
     },
     {
       "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": 27983.40468436003,
-      "required_pages_per_second_upper_estimate": 0.0017040425276998073
+      "required_token_positions_per_second": 28063.810579223056,
+      "required_pages_per_second_upper_estimate": 0.001708938824839838
     },
     {
       "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": 27983.40468436003,
-      "required_pages_per_second_upper_estimate": 0.002154239955624055
+      "required_token_positions_per_second": 28063.810579223056,
+      "required_pages_per_second_upper_estimate": 0.002160429823988374
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27983.40468436003,
-      "required_pages_per_second_upper_estimate": 0.34746023174888596
+      "required_token_positions_per_second": 28063.810579223056,
+      "required_pages_per_second_upper_estimate": 0.34845860386186545
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11430120917171394,
-    "producer_p95_seconds": 0.2712080999990576,
-    "gpu_update_seconds": 0.9880141573856633,
-    "producer_batches_per_second": 8.74881383361139,
-    "gpu_batches_per_second": 1.01213124581742,
-    "producer_headroom": 8.643951927939593,
+    "producer_batch_seconds": 0.08193009385931305,
+    "producer_p95_seconds": 0.28395018500305014,
+    "gpu_update_seconds": 0.9851833884764424,
+    "producer_batches_per_second": 12.205527333059797,
+    "gpu_batches_per_second": 1.0150394451397227,
+    "producer_headroom": 12.024682776125685,
     "sustainable": true,
-    "buffer_seconds": 60.268863600525464,
+    "buffer_seconds": 60.09618669706298,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 553342.366459169,
-    "upload_bytes_per_second": 5112855.071708027,
+    "generation_bytes_per_second": 554020.6045688958,
+    "upload_bytes_per_second": 5100479.052795358,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 118.73439069110205,
+    "estimated_full_shards_per_hour": 118.87992480087428,
     "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": 1.0609590189997107,
-      "bytes": 7221766,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 0.9498501529997156,
       "bytes": 6032509,
@@ -562,6 +556,12 @@
       "bytes": 1256653,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.3800626919983188,
+      "bytes": 7827681,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -242,3 +242,4 @@
 {"step": 64200, "loss": 0.6446213405579329, "lr": 0.0003, "grad_norm": 0.1374295009308241, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9395625453, "tokens_per_second": 150108.32928739092, "data_wait_s": 8.966002496890724e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 64300, "loss": 0.6748886741697788, "lr": 0.0003, "grad_norm": 0.1366752813543412, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9410240760, "tokens_per_second": 149030.21262801863, "data_wait_s": 1.0944000678136945e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 64400, "loss": 0.6146517023444176, "lr": 0.0003, "grad_norm": 0.1410616037631765, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9424838604, "tokens_per_second": 148309.56121770755, "data_wait_s": 9.709001460578293e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 64500, "loss": 0.6141596455127001, "lr": 0.0003, "grad_norm": 0.12958309178787517, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9439627986, "tokens_per_second": 152102.54236346792, "data_wait_s": 9.973999112844467e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bca3ed530994040abba0bab9262ae7212fed7190ea0e624ac5cb267ea8592f0f
-size 520089
+oid sha256:700882a958620bd1de9d9bb489ec0bd6745a09fe6ecee58ebf6e1fcd6184fe02
+size 522207