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": 76580.15228254475,
-      "required_pages_per_second_upper_estimate": 0.005297992625340705
+      "required_token_positions_per_second": 76793.1835215705,
+      "required_pages_per_second_upper_estimate": 0.005312730620757084
     },
     {
       "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": 9572.519035318093,
-      "required_pages_per_second_upper_estimate": 0.004007813823159191
+      "required_token_positions_per_second": 9599.147940196313,
+      "required_pages_per_second_upper_estimate": 0.004018962789556935
     },
     {
       "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": 19145.038070636187,
-      "required_pages_per_second_upper_estimate": 0.003987406541507307
+      "required_token_positions_per_second": 19198.295880392627,
+      "required_pages_per_second_upper_estimate": 0.003998498738776681
     },
     {
       "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": 28717.55710595428,
-      "required_pages_per_second_upper_estimate": 0.0017215736993312076
+      "required_token_positions_per_second": 28797.443820588942,
+      "required_pages_per_second_upper_estimate": 0.0017263627858936053
     },
     {
       "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": 28717.55710595428,
-      "required_pages_per_second_upper_estimate": 0.0022326637498036365
+      "required_token_positions_per_second": 28797.443820588942,
+      "required_pages_per_second_upper_estimate": 0.002238874590481959
     },
     {
       "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": 28717.55710595428,
-      "required_pages_per_second_upper_estimate": 0.36835798804471825
+      "required_token_positions_per_second": 28797.443820588942,
+      "required_pages_per_second_upper_estimate": 0.36938268904438043
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.25909494960163215,
-    "producer_p95_seconds": 0.2680201860021043,
-    "gpu_update_seconds": 0.9627559857543552,
-    "producer_batches_per_second": 3.8595889326964348,
-    "gpu_batches_per_second": 1.0386847911586472,
-    "producer_headroom": 3.7158423475047555,
+    "producer_batch_seconds": 0.40375336896102,
+    "producer_p95_seconds": 0.26272494300064864,
+    "gpu_update_seconds": 0.9600852135436014,
+    "producer_batches_per_second": 2.4767595192414213,
+    "gpu_batches_per_second": 1.0415742122608849,
+    "producer_headroom": 2.3779001919270475,
     "sustainable": true,
-    "buffer_seconds": 60.653627102524375,
+    "buffer_seconds": 60.48536845324689,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 407823.65282270935,
-    "upload_bytes_per_second": 4997486.297952912,
+    "generation_bytes_per_second": 364824.8149749637,
+    "upload_bytes_per_second": 4954087.929049629,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.50946224699936,
+    "estimated_full_shards_per_hour": 78.28291260658916,
     "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": 1.4825863509995543,
-      "bytes": 11209775,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 0.9785202229995775,
       "bytes": 2340420,
@@ -562,6 +556,12 @@
       "bytes": 11721944,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.162564084999758,
+      "bytes": 5892604,
+      "pages": 23,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -217,3 +217,4 @@
 {"step": 418800, "loss": 0.627915620803833, "lr": 0.0003, "grad_norm": 0.15717359423047464, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61301310935, "tokens_per_second": 151822.85662750484, "data_wait_s": 7.458998879883438e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 418900, "loss": 0.5589568614959717, "lr": 0.0003, "grad_norm": 0.1498780495751591, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61316001869, "tokens_per_second": 152837.25517190874, "data_wait_s": 1.4950001059332862e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 419000, "loss": 0.6161655187606812, "lr": 0.0003, "grad_norm": 0.17562282216041825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61330665018, "tokens_per_second": 152111.1166580027, "data_wait_s": 1.2300002708798274e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 419100, "loss": 0.5930498838424683, "lr": 0.0003, "grad_norm": 0.17148425860614736, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61345397171, "tokens_per_second": 152520.6746918214, "data_wait_s": 1.1173000530106947e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:63b97062b0426f6b004dd47fd2234206998ffafe62168ccbfd769cd30529ceaf
-size 467139
+oid sha256:d903cd68ce360ed380af5e70cd4e862f2517e79236e4fae2222e8becc7c2a3ab
+size 469257