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": 75902.7034466096,
-      "required_pages_per_second_upper_estimate": 0.006579053961108999
+      "required_token_positions_per_second": 76391.81220943856,
+      "required_pages_per_second_upper_estimate": 0.006621448669036189
     },
     {
       "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": 9487.8379308262,
-      "required_pages_per_second_upper_estimate": 0.0037848404064250037
+      "required_token_positions_per_second": 9548.97652617982,
+      "required_pages_per_second_upper_estimate": 0.0038092295062150234
     },
     {
       "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": 18975.6758616524,
-      "required_pages_per_second_upper_estimate": 0.004389688650104284
+      "required_token_positions_per_second": 19097.95305235964,
+      "required_pages_per_second_upper_estimate": 0.004417975326169349
     },
     {
       "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": 28463.513792478596,
-      "required_pages_per_second_upper_estimate": 0.0017332786534464113
+      "required_token_positions_per_second": 28646.92957853946,
+      "required_pages_per_second_upper_estimate": 0.0017444477125092596
     },
     {
       "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": 28463.513792478596,
-      "required_pages_per_second_upper_estimate": 0.0021912000837940953
+      "required_token_positions_per_second": 28646.92957853946,
+      "required_pages_per_second_upper_estimate": 0.0022053199387324527
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28463.513792478596,
-      "required_pages_per_second_upper_estimate": 0.3534215800498975
+      "required_token_positions_per_second": 28646.92957853946,
+      "required_pages_per_second_upper_estimate": 0.35569899025962554
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.14140635082040376,
-    "producer_p95_seconds": 0.3715155579993734,
-    "gpu_update_seconds": 0.9713488011907337,
-    "producer_batches_per_second": 7.071818162326188,
-    "gpu_batches_per_second": 1.029496303258051,
-    "producer_headroom": 6.8692020942144,
+    "producer_batch_seconds": 0.2782428986638479,
+    "producer_p95_seconds": 0.3441955170019355,
+    "gpu_update_seconds": 0.9651296109832379,
+    "producer_batches_per_second": 3.593982109883511,
+    "gpu_batches_per_second": 1.0361302654274978,
+    "producer_headroom": 3.4686585555925897,
     "sustainable": true,
-    "buffer_seconds": 60.223625673825495,
+    "buffer_seconds": 60.803165491943986,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 167,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 578192.5773328955,
-    "upload_bytes_per_second": 5246129.981460794,
+    "generation_bytes_per_second": 546386.313833311,
+    "upload_bytes_per_second": 5259046.753397099,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.033333730650654,
+    "estimated_full_shards_per_hour": 117.24178372621056,
     "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": 1.1093607039983908,
-      "bytes": 7338185,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 0.9098767819996283,
       "bytes": 5901703,
@@ -562,6 +556,12 @@
       "bytes": 5924124,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.1032993179978803,
+      "bytes": 8345826,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -312,3 +312,4 @@
 {"step": 146200, "loss": 0.616436455398798, "lr": 0.0003, "grad_norm": 0.1636305480061172, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21396067541, "tokens_per_second": 150013.2628389867, "data_wait_s": 1.0538999049458653e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 146300, "loss": 0.5969559364020824, "lr": 0.0003, "grad_norm": 0.16884647873112507, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21410563663, "tokens_per_second": 147842.31492261676, "data_wait_s": 1.2511001841630787e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 146400, "loss": 0.5712147578597069, "lr": 0.0003, "grad_norm": 0.1339102732111261, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21425080478, "tokens_per_second": 149677.5122500821, "data_wait_s": 1.1968000762863085e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 146500, "loss": 0.604755088686943, "lr": 0.0003, "grad_norm": 0.14296582557265414, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21439675555, "tokens_per_second": 149319.34816546435, "data_wait_s": 9.305997082265094e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2f38f129412741bf0516b7d858a95f9d38f9fd4a4e11124b1af48e9c02f224e3
-size 669915
+oid sha256:171e65f7b3d7e0317fc7547eb159f2ed265c615e98e9b3cdb2b43ddf75cc4bea
+size 672033