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": 77100.7867627799,
-      "required_pages_per_second_upper_estimate": 0.005334011326721803
+      "required_token_positions_per_second": 75745.17404160122,
+      "required_pages_per_second_upper_estimate": 0.005240226893215789
     },
     {
       "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": 9637.598345347487,
-      "required_pages_per_second_upper_estimate": 0.0040350611712579665
+      "required_token_positions_per_second": 9468.146755200152,
+      "required_pages_per_second_upper_estimate": 0.003964115329014861
     },
     {
       "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": 19275.196690694975,
-      "required_pages_per_second_upper_estimate": 0.004014515149551915
+      "required_token_positions_per_second": 18936.293510400305,
+      "required_pages_per_second_upper_estimate": 0.0039439305545744185
     },
     {
       "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": 28912.795036042462,
-      "required_pages_per_second_upper_estimate": 0.001733277915128936
+      "required_token_positions_per_second": 28404.44026560046,
+      "required_pages_per_second_upper_estimate": 0.0017028028228537283
     },
     {
       "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": 28912.795036042462,
-      "required_pages_per_second_upper_estimate": 0.002247842640107095
+      "required_token_positions_per_second": 28404.44026560046,
+      "required_pages_per_second_upper_estimate": 0.0022083202927215546
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28912.795036042462,
-      "required_pages_per_second_upper_estimate": 0.3708622905817327
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28404.44026560046,
+      "required_pages_per_second_upper_estimate": 0.364341661415329
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08189532783606523,
-    "producer_p95_seconds": 0.27555089499946916,
-    "gpu_update_seconds": 0.9562548333889623,
-    "producer_batches_per_second": 12.210708796498864,
-    "gpu_batches_per_second": 1.0457463482364895,
-    "producer_headroom": 11.676549305757158,
+    "producer_batch_seconds": 0.12006588582005406,
+    "producer_p95_seconds": 0.27644926499851863,
+    "gpu_update_seconds": 0.9733689430762501,
+    "producer_batches_per_second": 8.328760439903192,
+    "gpu_batches_per_second": 1.02735967395835,
+    "producer_headroom": 8.106956746523855,
     "sustainable": true,
-    "buffer_seconds": 60.24405450350463,
+    "buffer_seconds": 60.34887447072751,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 344028.5098457548,
-    "upload_bytes_per_second": 5996899.902374868,
+    "generation_bytes_per_second": 353908.61669902015,
+    "upload_bytes_per_second": 5997630.877891791,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.82050963906748,
+    "estimated_full_shards_per_hour": 75.9405505726619,
     "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": 1.2034739760019875,
-      "bytes": 5697150,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 0.9529857010020351,
       "bytes": 6624046,
@@ -562,6 +556,12 @@
       "bytes": 6924644,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.2151130479978747,
+      "bytes": 5819277,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -285,3 +285,4 @@
 {"step": 425600, "loss": 0.6000295877456665, "lr": 0.0003, "grad_norm": 0.1545670994982086, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62296119078, "tokens_per_second": 151801.25339343923, "data_wait_s": 8.053997589740902e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 425700, "loss": 0.6632568836212158, "lr": 0.0003, "grad_norm": 0.17485261796040918, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62310857885, "tokens_per_second": 152475.5499211867, "data_wait_s": 8.650000381749123e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 425800, "loss": 0.60506272315979, "lr": 0.0003, "grad_norm": 0.16878657541124026, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62325561358, "tokens_per_second": 151809.75710584078, "data_wait_s": 1.8686998373595998e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 425900, "loss": 0.6460074782371521, "lr": 0.0003, "grad_norm": 0.18997523605965633, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62340106425, "tokens_per_second": 151483.48869413583, "data_wait_s": 1.0556999768596143e-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:f682054a74a165cdeedef476aaea1228887a333ee3fa5d3b9c964bccf493b946
-size 612729
+oid sha256:13319016a5bb7233bc7c69b86f61024b1ae42a49b9ec94ffa966858a44ff8e01
+size 614847