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": 73697.04591404747,
-      "required_pages_per_second_upper_estimate": 0.006387873156374423
+      "required_token_positions_per_second": 75318.94044377226,
+      "required_pages_per_second_upper_estimate": 0.0065284548635568635
     },
     {
       "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": 9212.130739255934,
-      "required_pages_per_second_upper_estimate": 0.003674856685517765
+      "required_token_positions_per_second": 9414.867555471532,
+      "required_pages_per_second_upper_estimate": 0.0037557314326916915
     },
     {
       "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": 18424.261478511868,
-      "required_pages_per_second_upper_estimate": 0.004262128637126414
+      "required_token_positions_per_second": 18829.735110943064,
+      "required_pages_per_second_upper_estimate": 0.004355927825897168
     },
     {
       "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": 27636.392217767803,
-      "required_pages_per_second_upper_estimate": 0.0016829112891180468
+      "required_token_positions_per_second": 28244.602666414594,
+      "required_pages_per_second_upper_estimate": 0.0017199481144070283
     },
     {
       "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": 27636.392217767803,
-      "required_pages_per_second_upper_estimate": 0.0021275259753537976
+      "required_token_positions_per_second": 28244.602666414594,
+      "required_pages_per_second_upper_estimate": 0.0021743476993249067
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27636.392217767803,
-      "required_pages_per_second_upper_estimate": 0.34315149828982705
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28244.602666414594,
+      "required_pages_per_second_upper_estimate": 0.35070343651259167
     }
   ],
   "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": 60,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.4546642384766102,
-    "producer_p95_seconds": 0.2729480819998571,
-    "gpu_update_seconds": 1.0004200180016527,
-    "producer_batches_per_second": 2.199425235973214,
-    "gpu_batches_per_second": 0.9995801583394026,
-    "producer_headroom": 2.200349034165612,
+    "producer_batch_seconds": 0.061822844218724526,
+    "producer_p95_seconds": 0.20437664100063557,
+    "gpu_update_seconds": 0.9788772859203996,
+    "producer_batches_per_second": 16.175250631660944,
+    "gpu_batches_per_second": 1.021578510793352,
+    "producer_headroom": 15.833585437402494,
     "sustainable": true,
-    "buffer_seconds": 60.025201080099166,
+    "buffer_seconds": 60.69039172706477,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 364484.79055826424,
-    "upload_bytes_per_second": 5350695.617726562,
+    "generation_bytes_per_second": 364123.6995039226,
+    "upload_bytes_per_second": 5292114.322407744,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.20995128212877,
+    "estimated_full_shards_per_hour": 78.13246954763659,
     "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": 1.587895904999641,
-      "bytes": 7946846,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.0177612440002122,
       "bytes": 6784132,
@@ -562,6 +556,12 @@
       "bytes": 1181038,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 3.174646825998934,
+      "bytes": 11736199,
+      "pages": 33,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -117,3 +117,4 @@
 {"step": 166700, "loss": 0.6023055128753185, "lr": 0.0003, "grad_norm": 0.16025117601346572, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24397533861, "tokens_per_second": 147846.85166529688, "data_wait_s": 1.3814000340062194e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 166800, "loss": 0.58776630833745, "lr": 0.0003, "grad_norm": 0.13998994360880412, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24412049205, "tokens_per_second": 147603.0305860745, "data_wait_s": 1.4315000953502022e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 166900, "loss": 0.6116734612733126, "lr": 0.0003, "grad_norm": 0.14229868911171528, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24426748624, "tokens_per_second": 150449.8063104413, "data_wait_s": 1.085700023395475e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 167000, "loss": 0.5920278616249561, "lr": 0.0003, "grad_norm": 0.13092271103070144, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24441291686, "tokens_per_second": 148720.9678397098, "data_wait_s": 1.6636000509606674e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:74d584f66fcc28f7c78d6ccec1fd7c0d407db831c0002c7e6b16c2dbc3aa2525
-size 253773
+oid sha256:916cbce39a18302fae109342af858295658154a9de0524e36be1a45a26f29a10
+size 255891