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": 77421.94906637935,
-      "required_pages_per_second_upper_estimate": 0.005356230080084025
+      "required_token_positions_per_second": 76419.75808515011,
+      "required_pages_per_second_upper_estimate": 0.005286896182599132
     },
     {
       "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": 9677.74363329742,
-      "required_pages_per_second_upper_estimate": 0.004051869164993661
+      "required_token_positions_per_second": 9552.469760643764,
+      "required_pages_per_second_upper_estimate": 0.003999419610529514
     },
     {
       "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": 19355.48726659484,
-      "required_pages_per_second_upper_estimate": 0.00403123755910698
+      "required_token_positions_per_second": 19104.939521287528,
+      "required_pages_per_second_upper_estimate": 0.003979055071147839
     },
     {
       "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": 29033.230899892256,
-      "required_pages_per_second_upper_estimate": 0.001740497861271815
+      "required_token_positions_per_second": 28657.40928193129,
+      "required_pages_per_second_upper_estimate": 0.001717967929121439
     },
     {
       "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": 29033.230899892256,
-      "required_pages_per_second_upper_estimate": 0.002257205998780036
+      "required_token_positions_per_second": 28657.40928193129,
+      "required_pages_per_second_upper_estimate": 0.002227987521048167
     },
     {
       "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": 29033.230899892256,
-      "required_pages_per_second_upper_estimate": 0.37240711252924225
+      "required_token_positions_per_second": 28657.40928193129,
+      "required_pages_per_second_upper_estimate": 0.36758647633985314
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07396217163287133,
-    "producer_p95_seconds": 0.20793024900012824,
-    "gpu_update_seconds": 0.9522880899935461,
-    "producer_batches_per_second": 13.520425075722974,
-    "gpu_batches_per_second": 1.050102390765779,
-    "producer_headroom": 12.875339771261077,
+    "producer_batch_seconds": 0.15675093807034557,
+    "producer_p95_seconds": 0.21315120899998874,
+    "gpu_update_seconds": 0.9647766735645664,
+    "producer_batches_per_second": 6.3795471485550355,
+    "gpu_batches_per_second": 1.0365093056254084,
+    "producer_headroom": 6.154838276831242,
     "sustainable": true,
-    "buffer_seconds": 60.94643775958695,
+    "buffer_seconds": 60.780930434567686,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 232699.59856479696,
-    "upload_bytes_per_second": 5866517.117316585,
+    "generation_bytes_per_second": 252231.53830773654,
+    "upload_bytes_per_second": 5953983.2155751195,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 49.9319168825906,
+    "estimated_full_shards_per_hour": 54.12301647113869,
     "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": 1.0425156310002421,
-      "bytes": 44934,
-      "pages": 1,
-      "shards": 1
-    },
     {
       "seconds": 1.7656013979999443,
       "bytes": 11283965,
@@ -562,6 +556,12 @@
       "bytes": 12055069,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.2178526319985394,
+      "bytes": 7463726,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -81,3 +81,4 @@
 {"step": 493100, "loss": 0.582404375076294, "lr": 0.0003, "grad_norm": 0.1659191974143131, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72173501495, "tokens_per_second": 153322.7852714244, "data_wait_s": 1.0695999662857503e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 493200, "loss": 0.601795494556427, "lr": 0.0003, "grad_norm": 0.15764358765457287, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72188032382, "tokens_per_second": 151491.51400889465, "data_wait_s": 1.1154999810969457e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 493300, "loss": 0.5364515781402588, "lr": 0.0003, "grad_norm": 0.15361784279527516, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72202603504, "tokens_per_second": 152124.93001188117, "data_wait_s": 7.603999620187096e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 493400, "loss": 0.599118709564209, "lr": 0.0003, "grad_norm": 0.15573941228074265, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72217316920, "tokens_per_second": 154023.2147727918, "data_wait_s": 1.1088000974268652e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bf181af82bd4f46916d17894b3c262fca3bbd8820b61163f634c9b4a1f36d71e
-size 176742
+oid sha256:49b80260804b9c62421e435cf6fc9e179502ac24d990232670d5e684a829c711
+size 178860