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": 76278.4395037122,
-      "required_pages_per_second_upper_estimate": 0.006611621810244323
+      "required_token_positions_per_second": 75837.63856139279,
+      "required_pages_per_second_upper_estimate": 0.0065734143017638505
     },
     {
       "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": 9534.804937964025,
-      "required_pages_per_second_upper_estimate": 0.003803576247791617
+      "required_token_positions_per_second": 9479.704820174098,
+      "required_pages_per_second_upper_estimate": 0.00378159598698504
     },
     {
       "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": 19069.60987592805,
-      "required_pages_per_second_upper_estimate": 0.004411418631124768
+      "required_token_positions_per_second": 18959.409640348196,
+      "required_pages_per_second_upper_estimate": 0.004385925746081275
     },
     {
       "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": 28604.414813892075,
-      "required_pages_per_second_upper_estimate": 0.0017418587863999653
+      "required_token_positions_per_second": 28439.114460522294,
+      "required_pages_per_second_upper_estimate": 0.001731792862143674
     },
     {
       "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": 28604.414813892075,
-      "required_pages_per_second_upper_estimate": 0.002202047034461496
+      "required_token_positions_per_second": 28439.114460522294,
+      "required_pages_per_second_upper_estimate": 0.002189321755678426
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28604.414813892075,
-      "required_pages_per_second_upper_estimate": 0.3551710991704692
+      "required_token_positions_per_second": 28439.114460522294,
+      "required_pages_per_second_upper_estimate": 0.35311862200631133
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.11183638245307748,
-    "producer_p95_seconds": 0.3763003959975322,
-    "gpu_update_seconds": 0.9665640838970221,
-    "producer_batches_per_second": 8.941634001972158,
-    "gpu_batches_per_second": 1.034592549692277,
-    "producer_headroom": 8.642662277658683,
+    "producer_batch_seconds": 0.18019979966385336,
+    "producer_p95_seconds": 0.3579035799994017,
+    "gpu_update_seconds": 0.9721821696797037,
+    "producer_batches_per_second": 5.549395736651266,
+    "gpu_batches_per_second": 1.028613804272363,
+    "producer_headroom": 5.395023587668925,
     "sustainable": true,
-    "buffer_seconds": 60.893537285512394,
+    "buffer_seconds": 60.27529452014163,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 472609.3058780036,
-    "upload_bytes_per_second": 5566937.052532021,
+    "generation_bytes_per_second": 475952.776779387,
+    "upload_bytes_per_second": 5625832.0593866585,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.41095525984841,
+    "estimated_full_shards_per_hour": 102.12838628326614,
     "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": 0.9869726110009651,
-      "bytes": 2644359,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.3357893329994113,
       "bytes": 5825933,
@@ -562,6 +556,12 @@
       "bytes": 6184683,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.9518272349960171,
+      "bytes": 6809793,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -303,3 +303,4 @@
 {"step": 310500, "loss": 0.579857038334012, "lr": 0.0003, "grad_norm": 0.13863951613268993, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45449016212, "tokens_per_second": 148815.76590939896, "data_wait_s": 1.005100057227537e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 310600, "loss": 0.6004833858460188, "lr": 0.0003, "grad_norm": 0.14692711976080622, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45463602971, "tokens_per_second": 149616.04517196465, "data_wait_s": 1.473700103815645e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 310700, "loss": 0.5604292182251811, "lr": 0.0003, "grad_norm": 0.1467913101522624, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45478207863, "tokens_per_second": 149486.1072726009, "data_wait_s": 1.9542996597010642e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 310800, "loss": 0.6214486751705408, "lr": 0.0003, "grad_norm": 0.1710393720838941, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45492806850, "tokens_per_second": 149143.65493830875, "data_wait_s": 9.006995242089033e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1c2357cfb05e3aae2cca1df94eec607e5b725d6122628485fe78afd0baa96701
-size 650853
+oid sha256:7debf3055b119cdfda388436f94d3e133ac4fc93f45aa5d66b79be65f55d3c77
+size 652971