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": 75616.63497097934,
-      "required_pages_per_second_upper_estimate": 0.006554258270675292
+      "required_token_positions_per_second": 74036.766140725,
+      "required_pages_per_second_upper_estimate": 0.0064173192446097885
     },
     {
       "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": 9452.079371372418,
-      "required_pages_per_second_upper_estimate": 0.0037705757824207826
+      "required_token_positions_per_second": 9254.595767590625,
+      "required_pages_per_second_upper_estimate": 0.0036917966202292263
     },
     {
       "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": 18904.158742744836,
-      "required_pages_per_second_upper_estimate": 0.004373144423303324
+      "required_token_positions_per_second": 18509.19153518125,
+      "required_pages_per_second_upper_estimate": 0.004281775711018929
     },
     {
       "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": 28356.238114117252,
-      "required_pages_per_second_upper_estimate": 0.001726746127466192
+      "required_token_positions_per_second": 27763.787302771878,
+      "required_pages_per_second_upper_estimate": 0.00169066898140444
     },
     {
       "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": 28356.238114117252,
-      "required_pages_per_second_upper_estimate": 0.002182941705115755
+      "required_token_positions_per_second": 27763.787302771878,
+      "required_pages_per_second_upper_estimate": 0.002137333201649577
     },
     {
       "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": 28356.238114117252,
-      "required_pages_per_second_upper_estimate": 0.35208957515324946
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27763.787302771878,
+      "required_pages_per_second_upper_estimate": 0.34473331888165537
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.08161187750783938,
-    "producer_p95_seconds": 0.21608721499796957,
-    "gpu_update_seconds": 0.9750235517395852,
-    "producer_batches_per_second": 12.253118425120695,
-    "gpu_batches_per_second": 1.0256162512339861,
-    "producer_headroom": 11.947079046746934,
+    "producer_batch_seconds": 0.43840060327332253,
+    "producer_p95_seconds": 0.26046402100109844,
+    "gpu_update_seconds": 0.9958295566267965,
+    "producer_batches_per_second": 2.281018758946703,
+    "gpu_batches_per_second": 1.0041879088097467,
+    "producer_headroom": 2.2715058993793007,
     "sustainable": true,
-    "buffer_seconds": 60.45146020785428,
+    "buffer_seconds": 60.745602954234585,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 386602.5058368253,
-    "upload_bytes_per_second": 4791590.894190986,
+    "generation_bytes_per_second": 336504.296999693,
+    "upload_bytes_per_second": 4758085.566604582,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.95589810684747,
+    "estimated_full_shards_per_hour": 72.20598871701328,
     "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": 1.2685245129996474,
-      "bytes": 10318034,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 1.292125334000957,
       "bytes": 1081552,
@@ -562,6 +556,12 @@
       "bytes": 2354967,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.0578431400026602,
+      "bytes": 6390274,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -169,3 +169,4 @@
 {"step": 254000, "loss": 0.5805207639932632, "lr": 0.0003, "grad_norm": 0.14634467205067897, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37179037470, "tokens_per_second": 147658.0662841845, "data_wait_s": 1.003299985313788e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 254100, "loss": 0.5910433381795883, "lr": 0.0003, "grad_norm": 0.15330688472435477, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37193591900, "tokens_per_second": 148962.12180058032, "data_wait_s": 1.4354001905303448e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 254200, "loss": 0.632264755666256, "lr": 0.0003, "grad_norm": 0.13589899055627355, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37208166907, "tokens_per_second": 149553.803815343, "data_wait_s": 1.0056999599328265e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 254300, "loss": 0.5514777507632971, "lr": 0.0003, "grad_norm": 0.1366859609829462, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37222800816, "tokens_per_second": 135755.4666257701, "data_wait_s": 9.111001418204978e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b1ec6af68a905174ec827fdc43c213f2e009893faca75c27098f35f8d56cb65c
-size 364692
+oid sha256:bb09b60f64dcd5e10086806c798c6bfd5c94b39d2a1ca53b0936752e67fd2f41
+size 366810