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": 75371.9017255788,
-      "required_pages_per_second_upper_estimate": 0.006533045413234713
+      "required_token_positions_per_second": 72515.35055862638,
+      "required_pages_per_second_upper_estimate": 0.006285446797946019
     },
     {
       "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": 9421.48771569735,
-      "required_pages_per_second_upper_estimate": 0.0037583723135859864
+      "required_token_positions_per_second": 9064.418819828297,
+      "required_pages_per_second_upper_estimate": 0.0036159321923680777
     },
     {
       "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": 18842.9754313947,
-      "required_pages_per_second_upper_estimate": 0.004358990741011973
+      "required_token_positions_per_second": 18128.837639656595,
+      "required_pages_per_second_upper_estimate": 0.004193787531289233
     },
     {
       "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": 28264.46314709205,
-      "required_pages_per_second_upper_estimate": 0.0017211575134803965
+      "required_token_positions_per_second": 27193.256459484892,
+      "required_pages_per_second_upper_estimate": 0.0016559266463922594
     },
     {
       "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": 28264.46314709205,
-      "required_pages_per_second_upper_estimate": 0.0021758766141047868
+      "required_token_positions_per_second": 27193.256459484892,
+      "required_pages_per_second_upper_estimate": 0.002093412158002889
     },
     {
       "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": 28264.46314709205,
-      "required_pages_per_second_upper_estimate": 0.35095003721385265
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27193.256459484892,
+      "required_pages_per_second_upper_estimate": 0.3376492352519326
     }
   ],
   "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": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09757554533589996,
-    "producer_p95_seconds": 0.28143212699797004,
-    "gpu_update_seconds": 0.9781894620151144,
-    "producer_batches_per_second": 10.248469496711902,
-    "gpu_batches_per_second": 1.0222968441511882,
-    "producer_headroom": 10.024944863466924,
+    "producer_batch_seconds": 0.1752556488593484,
+    "producer_p95_seconds": 0.30708439299996826,
+    "gpu_update_seconds": 1.0167226584720601,
+    "producer_batches_per_second": 5.705950173409537,
+    "gpu_batches_per_second": 0.9835523893042857,
+    "producer_headroom": 5.801368829418057,
     "sustainable": true,
-    "buffer_seconds": 60.647746644937094,
+    "buffer_seconds": 61.003359508323605,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 296129.59133792704,
-    "upload_bytes_per_second": 5839316.734132429,
+    "generation_bytes_per_second": 328602.33103071793,
+    "upload_bytes_per_second": 5941130.540835884,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.542516757043444,
+    "estimated_full_shards_per_hour": 70.51041076842455,
     "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": 1.1912384690003819,
-      "bytes": 6120903,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.0073837069994624,
       "bytes": 6037950,
@@ -562,6 +556,12 @@
       "bytes": 6024956,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.1858630629976687,
+      "bytes": 13358845,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -169,3 +169,4 @@
 {"step": 131900, "loss": 0.5877761356532574, "lr": 0.0003, "grad_norm": 0.13578825804121405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19302738932, "tokens_per_second": 148233.61961941043, "data_wait_s": 1.4311997801996768e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 132000, "loss": 0.5825902055948973, "lr": 0.0003, "grad_norm": 0.14820855424023865, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19317194857, "tokens_per_second": 147764.81229459238, "data_wait_s": 1.3799999578623101e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 132100, "loss": 0.6819533929228783, "lr": 0.0003, "grad_norm": 0.1560970733064769, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19331878226, "tokens_per_second": 148650.67245640198, "data_wait_s": 8.839000656735152e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 132200, "loss": 0.5880027711391449, "lr": 0.0003, "grad_norm": 0.12591463884297227, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19346583756, "tokens_per_second": 136815.05019313606, "data_wait_s": 1.4776996977161616e-05, "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 sha256:6f6fbda0be04ce8a0d1c5cd72d5e9f01a84c8c23607f4582c384f39146a0acd8
-size 364692
+oid sha256:f92be1d8d9ef2635806ab37d5e2ca7006bfcb1a1c55d557bc9c4267aa7851753
+size 366810