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": 75027.15914736522,
-      "required_pages_per_second_upper_estimate": 0.006503164000297237
+      "required_token_positions_per_second": 74965.91349083069,
+      "required_pages_per_second_upper_estimate": 0.00649785538201452
     },
     {
       "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": 9378.394893420653,
-      "required_pages_per_second_upper_estimate": 0.003741181942484703
+      "required_token_positions_per_second": 9370.739186353836,
+      "required_pages_per_second_upper_estimate": 0.0037381279664727286
     },
     {
       "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": 18756.789786841306,
-      "required_pages_per_second_upper_estimate": 0.0043390532089600834
+      "required_token_positions_per_second": 18741.478372707672,
+      "required_pages_per_second_upper_estimate": 0.004335511182771952
     },
     {
       "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": 28135.184680261962,
-      "required_pages_per_second_upper_estimate": 0.00171328513312214
+      "required_token_positions_per_second": 28112.21755906151,
+      "required_pages_per_second_upper_estimate": 0.0017118865559402048
     },
     {
       "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": 28135.184680261962,
-      "required_pages_per_second_upper_estimate": 0.002165924399862506
+      "required_token_positions_per_second": 28112.21755906151,
+      "required_pages_per_second_upper_estimate": 0.0021641563272954347
     },
     {
       "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": 28135.184680261962,
-      "required_pages_per_second_upper_estimate": 0.34934483132301875
+      "required_token_positions_per_second": 28112.21755906151,
+      "required_pages_per_second_upper_estimate": 0.3490596565437192
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.10674087425778112,
-    "producer_p95_seconds": 0.2279312939990632,
-    "gpu_update_seconds": 0.9826841484853042,
-    "producer_batches_per_second": 9.368482382718565,
-    "gpu_batches_per_second": 1.0176209736784563,
-    "producer_headroom": 9.206259132861367,
+    "producer_batch_seconds": 0.08013641455465859,
+    "producer_p95_seconds": 0.23225831299896527,
+    "gpu_update_seconds": 0.9834869818403252,
+    "producer_batches_per_second": 12.47872150953211,
+    "gpu_batches_per_second": 1.0167902762970744,
+    "producer_headroom": 12.27266015463568,
     "sustainable": true,
-    "buffer_seconds": 60.92641720608886,
+    "buffer_seconds": 60.976192874100164,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 416406.6460923311,
-    "upload_bytes_per_second": 6091207.816030441,
+    "generation_bytes_per_second": 488942.4424791797,
+    "upload_bytes_per_second": 6089034.86801216,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.3511728008027,
+    "estimated_full_shards_per_hour": 104.91566615850013,
     "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": 1.0328473609997673,
-      "bytes": 7480746,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 1.141071416000159,
       "bytes": 10868715,
@@ -562,6 +556,12 @@
       "bytes": 1000268,
       "pages": 30,
       "shards": 1
+    },
+    {
+      "seconds": 1.0672549959999742,
+      "bytes": 7538580,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -118,3 +118,4 @@
 {"step": 51800, "loss": 0.5903803128749132, "lr": 0.0003, "grad_norm": 0.1342751621928546, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7581571065, "tokens_per_second": 147936.7649757513, "data_wait_s": 1.055899883795064e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 51900, "loss": 0.6394411474466324, "lr": 0.0003, "grad_norm": 0.1356565594248017, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7596081523, "tokens_per_second": 148360.54984097305, "data_wait_s": 7.036000170046464e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 52000, "loss": 0.6502221059054136, "lr": 0.0003, "grad_norm": 0.15132006725584615, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7610525400, "tokens_per_second": 147422.74991536338, "data_wait_s": 9.97200004348997e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 52100, "loss": 0.5977641493082047, "lr": 0.0003, "grad_norm": 0.1296948953688188, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7625174466, "tokens_per_second": 150817.8264087898, "data_wait_s": 1.3049999324721284e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:368e7a7327c95de98ea9f3dcd0fa94e05c9ab406235b7a9587e4143f2046f1c9
-size 255891
+oid sha256:cffe5ea43490ffe0cbc8232cb08511d66be8e9f3c111f1e2bd5d59e566ec1336
+size 258009