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": 76168.76457394392,
-      "required_pages_per_second_upper_estimate": 0.006602115465300576
+      "required_token_positions_per_second": 76250.04832826408,
+      "required_pages_per_second_upper_estimate": 0.006609160935113256
     },
     {
       "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": 9521.09557174299,
-      "required_pages_per_second_upper_estimate": 0.0037981073766327547
+      "required_token_positions_per_second": 9531.25604103301,
+      "required_pages_per_second_upper_estimate": 0.0038021605397450973
     },
     {
       "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": 19042.19114348598,
-      "required_pages_per_second_upper_estimate": 0.004405075789927505
+      "required_token_positions_per_second": 19062.51208206602,
+      "required_pages_per_second_upper_estimate": 0.004409776681431693
     },
     {
       "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": 28563.28671522897,
-      "required_pages_per_second_upper_estimate": 0.0017393543009738392
+      "required_token_positions_per_second": 28593.76812309903,
+      "required_pages_per_second_upper_estimate": 0.0017412104587895385
     },
     {
       "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": 28563.28671522897,
-      "required_pages_per_second_upper_estimate": 0.002198880879576546
+      "required_token_positions_per_second": 28593.76812309903,
+      "required_pages_per_second_upper_estimate": 0.002201227422732321
     },
     {
       "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": 28563.28671522897,
-      "required_pages_per_second_upper_estimate": 0.35466042583196505
+      "required_token_positions_per_second": 28593.76812309903,
+      "required_pages_per_second_upper_estimate": 0.3550389029030015
     }
   ],
   "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": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.08075632986697201,
-    "producer_p95_seconds": 0.21146970500194584,
-    "gpu_update_seconds": 0.9679558335021379,
-    "producer_batches_per_second": 12.38293025013985,
-    "gpu_batches_per_second": 1.0331049882533627,
-    "producer_headroom": 11.986129571472956,
+    "producer_batch_seconds": 0.09746381313286179,
+    "producer_p95_seconds": 0.2053375249997771,
+    "gpu_update_seconds": 0.9669239773146581,
+    "producer_batches_per_second": 10.260218309300182,
+    "gpu_batches_per_second": 1.0342074697301444,
+    "producer_headroom": 9.920851095745208,
     "sustainable": true,
-    "buffer_seconds": 60.01326167713255,
+    "buffer_seconds": 60.91621057082346,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 388018.9247684494,
-    "upload_bytes_per_second": 3966952.9770159055,
+    "generation_bytes_per_second": 419236.08439741394,
+    "upload_bytes_per_second": 4011209.54167114,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.2598286370288,
+    "estimated_full_shards_per_hour": 89.95830439512075,
     "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": 0.9624441889991431,
-      "bytes": 1027701,
-      "pages": 30,
-      "shards": 1
-    },
     {
       "seconds": 1.7364580140001635,
       "bytes": 17104134,
@@ -562,6 +556,12 @@
       "bytes": 1153413,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.3497758060002525,
+      "bytes": 7313270,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -119,3 +119,4 @@
 {"step": 292100, "loss": 0.5817364305257797, "lr": 0.0003, "grad_norm": 0.15640337190866618, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42755006915, "tokens_per_second": 151096.2911353824, "data_wait_s": 9.278999641537666e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 292200, "loss": 0.5701478384435177, "lr": 0.0003, "grad_norm": 0.15492571634711907, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42769727323, "tokens_per_second": 150911.93648607002, "data_wait_s": 9.900999430101365e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 292300, "loss": 0.5667089037597179, "lr": 0.0003, "grad_norm": 0.1332720000520859, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42784383346, "tokens_per_second": 150817.73813670804, "data_wait_s": 1.383500057272613e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 292400, "loss": 0.6011854503303766, "lr": 0.0003, "grad_norm": 0.15272123385569342, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42799047862, "tokens_per_second": 150602.1192327565, "data_wait_s": 1.3310997019289061e-05, "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 sha256:8d0ba797502143bd4caef2d6c1fe7d4f31c34a3a6bf91a21d558277b664938ef
-size 258009
+oid sha256:d47258330415b62cfc98b4d2c9eceda44bd1d63a1ca5b6b4b5613001a672a2ce
+size 260127