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": 75779.36463741628,
-      "required_pages_per_second_upper_estimate": 0.006568363265727477
+      "required_token_positions_per_second": 75379.69748973587,
+      "required_pages_per_second_upper_estimate": 0.0065337211303136645
     },
     {
       "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": 9472.420579677035,
-      "required_pages_per_second_upper_estimate": 0.0037786901945416605
+      "required_token_positions_per_second": 9422.462186216984,
+      "required_pages_per_second_upper_estimate": 0.0037587610444459007
     },
     {
       "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": 18944.84115935407,
-      "required_pages_per_second_upper_estimate": 0.004382555584399791
+      "required_token_positions_per_second": 18844.92437243397,
+      "required_pages_per_second_upper_estimate": 0.004359441594221217
     },
     {
       "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": 28417.261739031106,
-      "required_pages_per_second_upper_estimate": 0.0017304621460572296
+      "required_token_positions_per_second": 28267.386558650953,
+      "required_pages_per_second_upper_estimate": 0.0017213355339063786
     },
     {
       "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": 28417.261739031106,
-      "required_pages_per_second_upper_estimate": 0.002187639472685829
+      "required_token_positions_per_second": 28267.386558650953,
+      "required_pages_per_second_upper_estimate": 0.002176101666419112
     },
     {
       "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": 28417.261739031106,
-      "required_pages_per_second_upper_estimate": 0.35284728434174484
+      "required_token_positions_per_second": 28267.386558650953,
+      "required_pages_per_second_upper_estimate": 0.3509863362013851
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.12114809997672182,
-    "producer_p95_seconds": 0.39585821400396526,
-    "gpu_update_seconds": 0.9729297725412254,
-    "producer_batches_per_second": 8.254359748045133,
-    "gpu_batches_per_second": 1.0278234135934283,
-    "producer_headroom": 8.030912352138996,
+    "producer_batch_seconds": 0.11882231328104353,
+    "producer_p95_seconds": 0.3946625379976467,
+    "gpu_update_seconds": 0.9780882977149016,
+    "producer_batches_per_second": 8.415927719188213,
+    "gpu_batches_per_second": 1.0224025809697248,
+    "producer_headroom": 8.231520416552453,
     "sustainable": true,
-    "buffer_seconds": 60.321645897555975,
+    "buffer_seconds": 60.6414744583239,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 303111.03855716076,
-    "upload_bytes_per_second": 5265408.898065738,
+    "generation_bytes_per_second": 289461.8352417593,
+    "upload_bytes_per_second": 5241359.239888717,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.04057281051747,
+    "estimated_full_shards_per_hour": 62.11177151622376,
     "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": 1.0578519230002712,
-      "bytes": 8294494,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.1691894300020067,
       "bytes": 6431968,
@@ -562,6 +556,12 @@
       "bytes": 13201899,
       "pages": 33,
       "shards": 4
+    },
+    {
+      "seconds": 1.000070736001362,
+      "bytes": 6089030,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -361,3 +361,4 @@
 {"step": 391100, "loss": 0.577182162553072, "lr": 0.0003, "grad_norm": 0.14394163694856432, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57246223682, "tokens_per_second": 127790.04027333723, "data_wait_s": 1.1144002201035619e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 391200, "loss": 0.6033383291214705, "lr": 0.0003, "grad_norm": 0.19694065875609695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57260792556, "tokens_per_second": 148092.95161538015, "data_wait_s": 1.3568998838309199e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 391300, "loss": 0.58982796035707, "lr": 0.0003, "grad_norm": 0.14220308563823453, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57275549122, "tokens_per_second": 150404.7049975993, "data_wait_s": 1.0481002391315997e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 391400, "loss": 0.6011860184371471, "lr": 0.0003, "grad_norm": 0.1481824306291748, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57290258808, "tokens_per_second": 149502.60229895407, "data_wait_s": 2.0356004824861884e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3fc72a0ea55984869920af3a872c66e58fca533589934263e56141fcb609f734
-size 774480
+oid sha256:b88208eb3a9f672a7768953eb74fc01c0d7a7293cbfe944d7703eefb3e13f7ab
+size 776598