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": 75005.8552633991,
-      "required_pages_per_second_upper_estimate": 0.0065013174336292535
+      "required_token_positions_per_second": 74797.27910941307,
+      "required_pages_per_second_upper_estimate": 0.006483238581233173
     },
     {
       "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": 9375.731907924888,
-      "required_pages_per_second_upper_estimate": 0.003740119637755261
+      "required_token_positions_per_second": 9349.659888676633,
+      "required_pages_per_second_upper_estimate": 0.0037297191194657067
     },
     {
       "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": 18751.463815849776,
-      "required_pages_per_second_upper_estimate": 0.004337821139305081
+      "required_token_positions_per_second": 18699.319777353267,
+      "required_pages_per_second_upper_estimate": 0.004325758533702648
     },
     {
       "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": 28127.195723774665,
-      "required_pages_per_second_upper_estimate": 0.0017127986475868788
+      "required_token_positions_per_second": 28048.9796660299,
+      "required_pages_per_second_upper_estimate": 0.0017080357000381627
     },
     {
       "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": 28127.195723774665,
-      "required_pages_per_second_upper_estimate": 0.0021653093878772666
+      "required_token_positions_per_second": 28048.9796660299,
+      "required_pages_per_second_upper_estimate": 0.0021592880992361693
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28127.195723774665,
-      "required_pages_per_second_upper_estimate": 0.3492456352207639
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28048.9796660299,
+      "required_pages_per_second_upper_estimate": 0.3482744535558799
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09719965904675121,
-    "producer_p95_seconds": 0.19899525199980417,
-    "gpu_update_seconds": 0.98296325988269,
-    "producer_batches_per_second": 10.28810193170553,
-    "gpu_batches_per_second": 1.017332021259211,
-    "producer_headroom": 10.112826212794667,
+    "producer_batch_seconds": 0.07614557871086447,
+    "producer_p95_seconds": 0.21577082800104108,
+    "gpu_update_seconds": 0.9857043047267945,
+    "producer_batches_per_second": 13.132738852732887,
+    "gpu_batches_per_second": 1.014503026115086,
+    "producer_headroom": 12.944997219991631,
     "sustainable": true,
-    "buffer_seconds": 60.943722112726775,
+    "buffer_seconds": 60.12796258833446,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 263558.0299264932,
-    "upload_bytes_per_second": 5182437.212805479,
+    "generation_bytes_per_second": 285902.5265316924,
+    "upload_bytes_per_second": 5154052.374368634,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.55341790529343,
+    "estimated_full_shards_per_hour": 61.34802672350959,
     "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": [
@@ -502,6 +502,12 @@
       "bytes": 11693360,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 0.8287107050000486,
+      "bytes": 2337614,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -94,3 +94,4 @@
 {"step": 164400, "loss": 0.6361832730472088, "lr": 0.0003, "grad_norm": 0.1413945679153279, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24060826028, "tokens_per_second": 139704.4779702605, "data_wait_s": 9.821998901315965e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 164500, "loss": 0.6397268287837505, "lr": 0.0003, "grad_norm": 0.149824915661865, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24075511690, "tokens_per_second": 150490.8393519308, "data_wait_s": 1.091800004360266e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 164600, "loss": 0.6286317072808743, "lr": 0.0003, "grad_norm": 0.14458268787375328, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24090227342, "tokens_per_second": 150338.0715140591, "data_wait_s": 1.2294000043766573e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 164700, "loss": 0.6369891166687012, "lr": 0.0003, "grad_norm": 0.1658915428984174, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24104947467, "tokens_per_second": 150312.88167741676, "data_wait_s": 1.4675999409519136e-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 sha256:8413a42dfff66fb5e0abfaf3985c3bfeb79e69770d1c1d28c0abc386b77e58ad
-size 204276
+oid sha256:d02df1e2555391b1d39f19efeab8e4a1ef3a7b12472db74fa36d9490599a566a
+size 206394