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": 75340.44722879169,
-      "required_pages_per_second_upper_estimate": 0.006530319017173899
+      "required_token_positions_per_second": 74777.02048987121,
+      "required_pages_per_second_upper_estimate": 0.006481482615436286
     },
     {
       "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": 9417.55590359896,
-      "required_pages_per_second_upper_estimate": 0.003756803854954109
+      "required_token_positions_per_second": 9347.127561233901,
+      "required_pages_per_second_upper_estimate": 0.0037287089361871316
     },
     {
       "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": 18835.11180719792,
-      "required_pages_per_second_upper_estimate": 0.004357171629949108
+      "required_token_positions_per_second": 18694.255122467803,
+      "required_pages_per_second_upper_estimate": 0.004324586914929777
     },
     {
       "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": 28252.667710796883,
-      "required_pages_per_second_upper_estimate": 0.0017204392332958922
+      "required_token_positions_per_second": 28041.382683701708,
+      "required_pages_per_second_upper_estimate": 0.0017075730836726617
     },
     {
       "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": 28252.667710796883,
-      "required_pages_per_second_upper_estimate": 0.002174968568766927
+      "required_token_positions_per_second": 28041.382683701708,
+      "required_pages_per_second_upper_estimate": 0.002158703262506752
     },
     {
       "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": 28252.667710796883,
-      "required_pages_per_second_upper_estimate": 0.35080357737185247
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28041.382683701708,
+      "required_pages_per_second_upper_estimate": 0.348180124460828
     }
   ],
   "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": 0.10456494309391928,
-    "producer_p95_seconds": 0.4139775340008782,
-    "gpu_update_seconds": 0.9785978542987002,
-    "producer_batches_per_second": 9.563434650385732,
-    "gpu_batches_per_second": 1.0218702152342622,
-    "producer_headroom": 9.358756628593317,
+    "producer_batch_seconds": 0.1450643286401032,
+    "producer_p95_seconds": 0.40732164200016996,
+    "gpu_update_seconds": 0.9859713521212936,
+    "producer_batches_per_second": 6.893493454762033,
+    "gpu_batches_per_second": 1.0142282510019425,
+    "producer_headroom": 6.796787062431009,
     "sustainable": true,
-    "buffer_seconds": 60.67306696651941,
+    "buffer_seconds": 60.14425247939891,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 303911.47314950684,
+    "generation_bytes_per_second": 326536.84097016894,
     "upload_bytes_per_second": 5794773.637208084,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.21232744087128,
+    "estimated_full_shards_per_hour": 70.0672046835785,
     "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": [

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -387,3 +387,4 @@
 {"step": 233700, "loss": 0.5399864874780178, "lr": 0.0003, "grad_norm": 0.1503727513628934, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34207228558, "tokens_per_second": 147259.34759513917, "data_wait_s": 8.631999662611634e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 233800, "loss": 0.5587761066854, "lr": 0.0003, "grad_norm": 0.1490171192827732, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34221764765, "tokens_per_second": 147533.8362568862, "data_wait_s": 1.1696000001393259e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 233900, "loss": 0.5574611648917198, "lr": 0.0003, "grad_norm": 0.14534793742231727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34236369768, "tokens_per_second": 148103.06027596357, "data_wait_s": 1.2780001270584762e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 234000, "loss": 0.6060496270656586, "lr": 0.0003, "grad_norm": 0.13822100157510223, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34251105315, "tokens_per_second": 149677.64584513608, "data_wait_s": 1.2583004718180746e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1b43696ebc8048b73af2d20971ce21137de6ec0674751c05f1383da18ad6ac3f
-size 829548
+oid sha256:a95f8a0a551caa6a8e2fc70613b082dd9acc3f22c7393fad90a6129a632d4591
+size 831666