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": 75191.30722225014,
-      "required_pages_per_second_upper_estimate": 0.006517391939398768
+      "required_token_positions_per_second": 74065.43567614604,
+      "required_pages_per_second_upper_estimate": 0.0064198042472778105
     },
     {
       "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": 9398.913402781267,
-      "required_pages_per_second_upper_estimate": 0.003749367082647705
+      "required_token_positions_per_second": 9258.179459518255,
+      "required_pages_per_second_upper_estimate": 0.003693226208520127
     },
     {
       "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": 18797.826805562534,
-      "required_pages_per_second_upper_estimate": 0.004348546401014377
+      "required_token_positions_per_second": 18516.35891903651,
+      "required_pages_per_second_upper_estimate": 0.004283433759132196
     },
     {
       "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": 28196.740208343803,
-      "required_pages_per_second_upper_estimate": 0.0017170335418254276
+      "required_token_positions_per_second": 27774.538378554764,
+      "required_pages_per_second_upper_estimate": 0.0016913236655130824
     },
     {
       "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": 28196.740208343803,
-      "required_pages_per_second_upper_estimate": 0.0021706631148108537
+      "required_token_positions_per_second": 27774.538378554764,
+      "required_pages_per_second_upper_estimate": 0.0021381608492242266
     },
     {
       "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": 28196.740208343803,
-      "required_pages_per_second_upper_estimate": 0.35010914496869516
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27774.538378554764,
+      "required_pages_per_second_upper_estimate": 0.3448668112613428
     }
   ],
   "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": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10645257582046952,
-    "producer_p95_seconds": 0.33711466700333403,
-    "gpu_update_seconds": 0.9805388777465339,
-    "producer_batches_per_second": 9.393854421019208,
-    "gpu_batches_per_second": 1.0198473744337313,
-    "producer_headroom": 9.21103947170049,
+    "producer_batch_seconds": 0.3481762640548709,
+    "producer_p95_seconds": 0.3698174349992769,
+    "gpu_update_seconds": 0.9954440870688793,
+    "producer_batches_per_second": 2.8721084784872204,
+    "gpu_batches_per_second": 1.004576764270644,
+    "producer_headroom": 2.8590234023304992,
     "sustainable": true,
-    "buffer_seconds": 60.793410420285106,
+    "buffer_seconds": 60.722089311201636,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 299476.5433274683,
-    "upload_bytes_per_second": 5973164.954195643,
+    "generation_bytes_per_second": 324767.7832537485,
+    "upload_bytes_per_second": 5969624.902411419,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.26069474094426,
+    "estimated_full_shards_per_hour": 69.68760607919064,
     "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": 0.9104454030020861,
-      "bytes": 6024195,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.20260026500182,
       "bytes": 1126978,
@@ -562,6 +556,12 @@
       "bytes": 6263311,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 0.9876477170037106,
+      "bytes": 6238360,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -266,3 +266,4 @@
 {"step": 111600, "loss": 0.6359089203178883, "lr": 0.0003, "grad_norm": 0.12527254653037218, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16330553847, "tokens_per_second": 148136.05730953, "data_wait_s": 1.4473000192083418e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 111700, "loss": 0.5992363654077053, "lr": 0.0003, "grad_norm": 0.13714246636962182, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16345018972, "tokens_per_second": 147390.0481111329, "data_wait_s": 1.2933000107295811e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 111800, "loss": 0.6815985608845949, "lr": 0.0003, "grad_norm": 0.21084688447450772, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16359614931, "tokens_per_second": 148037.53453362366, "data_wait_s": 1.5261000953614712e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 111900, "loss": 0.5819695480167866, "lr": 0.0003, "grad_norm": 0.14860150425731405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16374251401, "tokens_per_second": 149879.69115942155, "data_wait_s": 9.020004654303193e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b151f19f267cf6608dd13cb01d1f4a39cb9c6f89fd8502b2ade4ca358b9759d6
-size 571704
+oid sha256:f93542a2587b5c3785166820eb702291d3378e7792b47a0f57d58f22729ffdd1
+size 573822