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": 73599.0512386282,
-      "required_pages_per_second_upper_estimate": 0.006379379226274315
+      "required_token_positions_per_second": 74977.42425916213,
+      "required_pages_per_second_upper_estimate": 0.006498853106239714
     },
     {
       "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": 9199.881404828526,
-      "required_pages_per_second_upper_estimate": 0.003669970242870802
+      "required_token_positions_per_second": 9372.178032395266,
+      "required_pages_per_second_upper_estimate": 0.0037387019436713204
     },
     {
       "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": 18399.76280965705,
-      "required_pages_per_second_upper_estimate": 0.004256461301248698
+      "required_token_positions_per_second": 18744.356064790532,
+      "required_pages_per_second_upper_estimate": 0.004336176886189672
     },
     {
       "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": 27599.64421448558,
-      "required_pages_per_second_upper_estimate": 0.0016806735285200321
+      "required_token_positions_per_second": 28116.5340971858,
+      "required_pages_per_second_upper_estimate": 0.0017121494104648486
     },
     {
       "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": 27599.64421448558,
-      "required_pages_per_second_upper_estimate": 0.002124697012336148
+      "required_token_positions_per_second": 28116.5340971858,
+      "required_pages_per_second_upper_estimate": 0.0021644886263491843
     },
     {
       "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": 27599.64421448558,
-      "required_pages_per_second_upper_estimate": 0.3426952110767173
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28116.5340971858,
+      "required_pages_per_second_upper_estimate": 0.34911325350069905
     }
   ],
   "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": 60,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.24510808823437458,
-    "producer_p95_seconds": 0.18048557800011622,
-    "gpu_update_seconds": 1.001752043799501,
-    "producer_batches_per_second": 4.079832726873504,
-    "gpu_batches_per_second": 0.9982510204892064,
-    "producer_headroom": 4.086980772505624,
+    "producer_batch_seconds": 0.05083539467187137,
+    "producer_p95_seconds": 0.1839790499998344,
+    "gpu_update_seconds": 0.983335993847382,
+    "producer_batches_per_second": 19.671333456831164,
+    "gpu_batches_per_second": 1.0169464010845557,
+    "producer_headroom": 19.34353023507633,
     "sustainable": true,
-    "buffer_seconds": 60.105122627970054,
+    "buffer_seconds": 60.96683161853768,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 418929.98940490134,
-    "upload_bytes_per_second": 4306177.841782511,
+    "generation_bytes_per_second": 417617.0856459662,
+    "upload_bytes_per_second": 4467626.819707898,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.89262353525429,
+    "estimated_full_shards_per_hour": 89.61090495142211,
     "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": [
@@ -238,6 +238,12 @@
       "bytes": 7946846,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.0177612440002122,
+      "bytes": 6784132,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -40,3 +40,4 @@
 {"step": 159000, "loss": 0.5832870006561279, "lr": 0.0003, "grad_norm": 0.14141191893168306, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23270696232, "tokens_per_second": 149508.56279893636, "data_wait_s": 1.2864999916928355e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 159100, "loss": 0.6049460880458355, "lr": 0.0003, "grad_norm": 0.13986737619423734, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23285302788, "tokens_per_second": 147677.8955632395, "data_wait_s": 1.3328000022738706e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 159200, "loss": 0.5842249542474747, "lr": 0.0003, "grad_norm": 0.14998375597947464, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23299799851, "tokens_per_second": 147627.9249082167, "data_wait_s": 1.2814000001526438e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 159300, "loss": 0.5889497511088848, "lr": 0.0003, "grad_norm": 0.14231023699674406, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23314483323, "tokens_per_second": 149607.7924737093, "data_wait_s": 1.3675999980478082e-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 sha256:247819779370b7cb4cdd0835443caeb5f69509dce7622a8cf7dcc53605cf5068
-size 89121
+oid sha256:528ebb94b306994607c883dfa0dbbf29e73bea8b7ffc85ff26dc301a84940c65
+size 91239