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": 69597.10018760353,
-      "required_pages_per_second_upper_estimate": 0.0060325002520237045
+      "required_token_positions_per_second": 75222.15893653342,
+      "required_pages_per_second_upper_estimate": 0.006520066087799889
     },
     {
       "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": 8699.637523450441,
-      "required_pages_per_second_upper_estimate": 0.0034704154792765444
+      "required_token_positions_per_second": 9402.769867066678,
+      "required_pages_per_second_upper_estimate": 0.003750905483910435
     },
     {
       "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": 17399.275046900882,
-      "required_pages_per_second_upper_estimate": 0.004025016065318295
+      "required_token_positions_per_second": 18805.539734133356,
+      "required_pages_per_second_upper_estimate": 0.004350330651296331
     },
     {
       "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": 26098.912570351327,
-      "required_pages_per_second_upper_estimate": 0.001589286845122679
+      "required_token_positions_per_second": 28208.309601200035,
+      "required_pages_per_second_upper_estimate": 0.001717738057150459
     },
     {
       "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": 26098.912570351327,
-      "required_pages_per_second_upper_estimate": 0.002009166536079616
+      "required_token_positions_per_second": 28208.309601200035,
+      "required_pages_per_second_upper_estimate": 0.0021715537586990555
     },
     {
       "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": 20,
-      "required_token_positions_per_second": 26098.912570351327,
-      "required_pages_per_second_upper_estimate": 0.32406114668228675
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28208.309601200035,
+      "required_pages_per_second_upper_estimate": 0.3502527981077025
     }
   ],
   "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": 57,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.04300748954687528,
-    "producer_p95_seconds": 0.1263296930000024,
-    "gpu_update_seconds": 1.0593544817422185,
-    "producer_batches_per_second": 23.25176406565343,
-    "gpu_batches_per_second": 0.9439710854438413,
-    "producer_headroom": 24.631860471362625,
+    "producer_batch_seconds": 0.06362290960937678,
+    "producer_p95_seconds": 0.12642385900005593,
+    "gpu_update_seconds": 0.9801367182535391,
+    "producer_batches_per_second": 15.717608737790568,
+    "gpu_batches_per_second": 1.0202658275897003,
+    "producer_headroom": 15.4054054470512,
     "sustainable": true,
-    "buffer_seconds": 60.38320545930645,
+    "buffer_seconds": 60.76847653171942,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 95,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 910173.0309263988,
-    "upload_bytes_per_second": 3293326.3783964273,
+    "generation_bytes_per_second": 957712.7056038796,
+    "upload_bytes_per_second": 3672933.616484457,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.65097234651553,
+    "estimated_full_shards_per_hour": 102.75142610591551,
     "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": [
@@ -208,6 +208,12 @@
       "bytes": 494707,
       "pages": 15,
       "shards": 1
+    },
+    {
+      "seconds": 0.9781999550000364,
+      "bytes": 5863999,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -1 +1,2 @@
 {"step": 100, "loss": 2.8035809621214867, "lr": 2.9999999999999997e-05, "grad_norm": 1.586405728010338, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14758037, "tokens_per_second": 46373.26368537362, "data_wait_s": 8.257999979832675e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 200, "loss": 2.3905331194400787, "lr": 5.9999999999999995e-05, "grad_norm": 1.9926730765732787, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29475318, "tokens_per_second": 151060.5326281325, "data_wait_s": 9.137000006376184e-06, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4430eb84436ec72df1c6721c0d95aac3150cb06c67c25b1cf73c5b969e6be77c
-size 2209
+oid sha256:f843ca14bf773d659916719a16d5ca618647f794dc385592e45c686782c30a2d
+size 4292