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": 74325.71132136564,
-      "required_pages_per_second_upper_estimate": 0.0064423642805428545
+      "required_token_positions_per_second": 75121.51712748398,
+      "required_pages_per_second_upper_estimate": 0.006511342710865816
     },
     {
       "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": 9290.713915170705,
-      "required_pages_per_second_upper_estimate": 0.0037062046893133497
+      "required_token_positions_per_second": 9390.189640935498,
+      "required_pages_per_second_upper_estimate": 0.0037458870436155645
     },
     {
       "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": 18581.42783034141,
-      "required_pages_per_second_upper_estimate": 0.004298486306588858
+      "required_token_positions_per_second": 18780.379281870995,
+      "required_pages_per_second_upper_estimate": 0.004344510223474269
     },
     {
       "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": 27872.141745512115,
-      "required_pages_per_second_upper_estimate": 0.0016972671984754946
+      "required_token_positions_per_second": 28170.56892280649,
+      "required_pages_per_second_upper_estimate": 0.0017154398478463285
     },
     {
       "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": 27872.141745512115,
-      "required_pages_per_second_upper_estimate": 0.0021456746265960157
+      "required_token_positions_per_second": 28170.56892280649,
+      "required_pages_per_second_upper_estimate": 0.002168648376803436
     },
     {
       "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": 27872.141745512115,
-      "required_pages_per_second_upper_estimate": 0.3460787184214971
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28170.56892280649,
+      "required_pages_per_second_upper_estimate": 0.34978418519197996
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.38446748673419506,
-    "producer_p95_seconds": 0.4180796000000555,
-    "gpu_update_seconds": 0.9919582159290573,
-    "producer_batches_per_second": 2.6010001742783486,
-    "gpu_batches_per_second": 1.0081069786426546,
-    "producer_headroom": 2.580083492508318,
+    "producer_batch_seconds": 0.15273342950814595,
+    "producer_p95_seconds": 0.39560157399682794,
+    "gpu_update_seconds": 0.9814498271498014,
+    "producer_batches_per_second": 6.547355108965622,
+    "gpu_batches_per_second": 1.0189007856917858,
+    "producer_headroom": 6.4259005399826785,
     "sustainable": true,
-    "buffer_seconds": 60.509451171672495,
+    "buffer_seconds": 60.84988928328769,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 399199.19359514734,
+    "generation_bytes_per_second": 381930.9938936197,
     "upload_bytes_per_second": 5771919.718944441,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.65885406390014,
+    "estimated_full_shards_per_hour": 81.95350039106792,
     "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

@@ -365,3 +365,4 @@
 {"step": 231500, "loss": 0.5812797918915749, "lr": 0.0003, "grad_norm": 0.22347676210351913, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33885191009, "tokens_per_second": 149381.63819913988, "data_wait_s": 1.534100010758266e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 231600, "loss": 0.6095680668950081, "lr": 0.0003, "grad_norm": 0.16434959095671053, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33899904451, "tokens_per_second": 149145.79090702962, "data_wait_s": 1.3511002180166543e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 231700, "loss": 0.583973566070199, "lr": 0.0003, "grad_norm": 0.16444283742634583, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33914520636, "tokens_per_second": 148426.13955550146, "data_wait_s": 1.5500998415518552e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 231800, "loss": 0.6330943591892719, "lr": 0.0003, "grad_norm": 0.15224543043698488, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33929256447, "tokens_per_second": 149345.46635311778, "data_wait_s": 9.500006854068488e-06, "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 sha256:70bb0808068d2712cbc7bdfa8cd752c9b31b9c8d87227d38e75fd3dc2ca0c248
-size 782952
+oid sha256:cd69d7489f60cd911193c32873ffbfae1d6d5ebce1b727c7be0b83256741a68b
+size 785070