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": 77497.54145697228,
-      "required_pages_per_second_upper_estimate": 0.005361459737063755
+      "required_token_positions_per_second": 76534.80309603289,
+      "required_pages_per_second_upper_estimate": 0.005294855263393205
     },
     {
       "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": 9687.192682121535,
-      "required_pages_per_second_upper_estimate": 0.004055825284417741
+      "required_token_positions_per_second": 9566.850387004111,
+      "required_pages_per_second_upper_estimate": 0.004005440478484964
     },
     {
       "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": 19374.38536424307,
-      "required_pages_per_second_upper_estimate": 0.004035173534470758
+      "required_token_positions_per_second": 19133.700774008223,
+      "required_pages_per_second_upper_estimate": 0.00398504528160432
     },
     {
       "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": 29061.578046364604,
-      "required_pages_per_second_upper_estimate": 0.0017421972294192458
+      "required_token_positions_per_second": 28700.551161012332,
+      "required_pages_per_second_upper_estimate": 0.0017205542188985124
     },
     {
       "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": 29061.578046364604,
-      "required_pages_per_second_upper_estimate": 0.002259409864732341
+      "required_token_positions_per_second": 28700.551161012332,
+      "required_pages_per_second_upper_estimate": 0.0022313416123856465
     },
     {
       "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": 29061.578046364604,
-      "required_pages_per_second_upper_estimate": 0.372770719287395
+      "required_token_positions_per_second": 28700.551161012332,
+      "required_pages_per_second_upper_estimate": 0.368139854042564
     }
   ],
   "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": 0.1396459998594537,
-    "producer_p95_seconds": 0.27141610000035143,
-    "gpu_update_seconds": 0.9513592123555922,
-    "producer_batches_per_second": 7.160964159420585,
-    "gpu_batches_per_second": 1.0511276781815901,
-    "producer_headroom": 6.812649222412993,
+    "producer_batch_seconds": 0.363472916593679,
+    "producer_p95_seconds": 0.2780603829996835,
+    "gpu_update_seconds": 0.9633264478055686,
+    "producer_batches_per_second": 2.7512366240973196,
+    "gpu_batches_per_second": 1.038069703450967,
+    "producer_headroom": 2.6503390041642554,
     "sustainable": true,
-    "buffer_seconds": 60.8869895907579,
+    "buffer_seconds": 60.68956621175082,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 461108.29557427915,
-    "upload_bytes_per_second": 5486491.221233017,
+    "generation_bytes_per_second": 401627.36138934706,
+    "upload_bytes_per_second": 5503704.79314713,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.94310617848664,
+    "estimated_full_shards_per_hour": 86.17988234768208,
     "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": 1.185240850001719,
-      "bytes": 5667842,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 0.9565514620007889,
       "bytes": 2665835,
@@ -562,6 +556,12 @@
       "bytes": 2607511,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 0.9523471370011976,
+      "bytes": 5723812,
+      "pages": 24,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -222,3 +222,4 @@
 {"step": 507200, "loss": 0.5564034581184387, "lr": 0.0003, "grad_norm": 0.16956993233946016, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74237249256, "tokens_per_second": 153147.2302347653, "data_wait_s": 7.195001671789214e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 507300, "loss": 0.5764183402061462, "lr": 0.0003, "grad_norm": 0.15764874691966407, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74251910837, "tokens_per_second": 153151.64194877763, "data_wait_s": 1.147299917647615e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 507400, "loss": 0.5982879996299744, "lr": 0.0003, "grad_norm": 0.1681867520757786, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74266776156, "tokens_per_second": 154267.11232209133, "data_wait_s": 1.3694003428099677e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 507500, "loss": 0.5969457626342773, "lr": 0.0003, "grad_norm": 0.14158622507568192, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74281256347, "tokens_per_second": 150296.1681045733, "data_wait_s": 1.5696998161729425e-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 sha256:339342a06aa55e9360eb1e3cf65a0b01afa0a6ba8abd606f0e3318c6260da3e9
-size 477729
+oid sha256:7c1be1cd2c180b7bb606af79008299a68911af5829cf90b72a798cf689886a33
+size 479847