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": 76712.36822549939,
-      "required_pages_per_second_upper_estimate": 0.005307139631057562
+      "required_token_positions_per_second": 76869.05410314113,
+      "required_pages_per_second_upper_estimate": 0.005317979523634143
     },
     {
       "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": 9589.046028187424,
-      "required_pages_per_second_upper_estimate": 0.004014733329950723
+      "required_token_positions_per_second": 9608.63176289264,
+      "required_pages_per_second_upper_estimate": 0.004022933468075148
     },
     {
       "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": 19178.092056374848,
-      "required_pages_per_second_upper_estimate": 0.003994290815044447
+      "required_token_positions_per_second": 19217.26352578528,
+      "required_pages_per_second_upper_estimate": 0.004002449199101524
     },
     {
       "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": 28767.13808456227,
-      "required_pages_per_second_upper_estimate": 0.0017245460032929859
+      "required_token_positions_per_second": 28825.895288677923,
+      "required_pages_per_second_upper_estimate": 0.0017280684079626634
     },
     {
       "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": 28767.13808456227,
-      "required_pages_per_second_upper_estimate": 0.0022365184528067303
+      "required_token_positions_per_second": 28825.895288677923,
+      "required_pages_per_second_upper_estimate": 0.002241086566981095
     },
     {
       "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": 28767.13808456227,
-      "required_pages_per_second_upper_estimate": 0.3689939596023944
+      "required_token_positions_per_second": 28825.895288677923,
+      "required_pages_per_second_upper_estimate": 0.3697476339282195
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08221072063288659,
+    "producer_batch_seconds": 0.0702126155156435,
     "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,
+    "gpu_update_seconds": 0.95913759913155,
+    "producer_batches_per_second": 14.242454759105193,
+    "gpu_batches_per_second": 1.0426032728833161,
+    "producer_headroom": 13.660473863387873,
     "sustainable": true,
-    "buffer_seconds": 60.54908885547918,
+    "buffer_seconds": 60.425668745287645,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 449633.7837180288,
+    "generation_bytes_per_second": 449160.5278017919,
     "upload_bytes_per_second": 5009302.415792151,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.48094304710052,
+    "estimated_full_shards_per_hour": 96.37939334430997,
     "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": [

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

@@ -78,3 +78,4 @@
 {"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}
+{"step": 405200, "loss": 0.6539797186851501, "lr": 0.0003, "grad_norm": 0.15852719614158323, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59311179550, "tokens_per_second": 152195.71173943108, "data_wait_s": 1.4334998923004605e-05, "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:d97f38434dce071373b2478e347f4252c58a1d13cb2ef1b4837cb6ccd9291590
-size 171171
+oid sha256:04aed5d8d1271e27a16e19a6c97d1e830049bb39ee9ff43f3e22acf400275507
+size 173289