Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76765.55632890557,
-      "required_pages_per_second_upper_estimate": 0.005310819307464605
+      "required_token_positions_per_second": 76712.36822549939,
+      "required_pages_per_second_upper_estimate": 0.005307139631057562
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9595.694541113196,
-      "required_pages_per_second_upper_estimate": 0.004017516923476006
+      "required_token_positions_per_second": 9589.046028187424,
+      "required_pages_per_second_upper_estimate": 0.004014733329950723
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19191.389082226393,
-      "required_pages_per_second_upper_estimate": 0.003997060234863171
+      "required_token_positions_per_second": 19178.092056374848,
+      "required_pages_per_second_upper_estimate": 0.003994290815044447
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28787.08362333959,
-      "required_pages_per_second_upper_estimate": 0.0017257417078876118
+      "required_token_positions_per_second": 28767.13808456227,
+      "required_pages_per_second_upper_estimate": 0.0017245460032929859
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28787.08362333959,
-      "required_pages_per_second_upper_estimate": 0.0022380691307155136
+      "required_token_positions_per_second": 28767.13808456227,
+      "required_pages_per_second_upper_estimate": 0.0022365184528067303
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28787.08362333959,
-      "required_pages_per_second_upper_estimate": 0.3692497995579789
+      "required_token_positions_per_second": 28767.13808456227,
+      "required_pages_per_second_upper_estimate": 0.3689939596023944
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,39 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.06708011378908907,
-    "producer_p95_seconds": 0.1769912210002076,
-    "gpu_update_seconds": 0.9604307390688211,
-    "producer_batches_per_second": 14.907547759149079,
-    "gpu_batches_per_second": 1.0411994944784284,
-    "producer_headroom": 14.3176671120233,
+    "producer_batch_seconds": 0.08221072063288659,
+    "producer_p95_seconds": 0.19439995900029317,
+    "gpu_update_seconds": 0.9610966484996695,
+    "producer_batches_per_second": 12.163863694438557,
+    "gpu_batches_per_second": 1.0404780846557535,
+    "producer_headroom": 11.690648629531704,
     "sustainable": true,
-    "buffer_seconds": 60.507136561335734,
+    "buffer_seconds": 60.54908885547918,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 432985.9895226683,
-    "upload_bytes_per_second": 4974423.660555432,
+    "generation_bytes_per_second": 449633.7837180288,
+    "upload_bytes_per_second": 5009302.415792151,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.90871395359075,
+    "estimated_full_shards_per_hour": 96.48094304710052,
     "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": 0.8249086619999844,
-      "bytes": 105910,
-      "pages": 3,
-      "shards": 1
-    },
-    {
-      "seconds": 1.08548185799998,
-      "bytes": 2013983,
-      "pages": 18,
-      "shards": 3
-    },
     {
       "seconds": 1.0607301400000324,
       "bytes": 873319,
@@ -562,6 +550,18 @@
       "bytes": 883918,
       "pages": 25,
       "shards": 2
+    },
+    {
+      "seconds": 1.464534527000069,
+      "bytes": 7593844,
+      "pages": 39,
+      "shards": 3
+    },
+    {
+      "seconds": 1.0184807959994941,
+      "bytes": 301114,
+      "pages": 8,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -77,3 +77,4 @@
 {"step": 404800, "loss": 0.624845027923584, "lr": 0.0003, "grad_norm": 0.16645235871608838, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59252363994, "tokens_per_second": 152852.3598016883, "data_wait_s": 1.3953000234323554e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 404900, "loss": 0.5621156692504883, "lr": 0.0003, "grad_norm": 0.1732311903236872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59267149727, "tokens_per_second": 154028.61356223238, "data_wait_s": 7.122999704733957e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 405000, "loss": 0.5751948356628418, "lr": 0.0003, "grad_norm": 0.1561981491970473, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59281882394, "tokens_per_second": 153122.1400274555, "data_wait_s": 1.2495000191847794e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 405100, "loss": 0.6122459769248962, "lr": 0.0003, "grad_norm": 0.1600886170270285, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59296549180, "tokens_per_second": 142635.64854577454, "data_wait_s": 7.482000000891276e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6c11e449210b4e53ac36c3814f31db431a7f6e5e46e036294a1eb3ea15c3c2fe
-size 169053
+oid sha256:d97f38434dce071373b2478e347f4252c58a1d13cb2ef1b4837cb6ccd9291590
+size 171171