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": 76947.87730255195,
-      "required_pages_per_second_upper_estimate": 0.005323432695464403
+      "required_token_positions_per_second": 77292.3550304763,
+      "required_pages_per_second_upper_estimate": 0.005347264463980661
     },
     {
       "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": 9618.484662818993,
-      "required_pages_per_second_upper_estimate": 0.004027058671522365
+      "required_token_positions_per_second": 9661.544378809538,
+      "required_pages_per_second_upper_estimate": 0.004045086875418486
     },
     {
       "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": 19236.969325637987,
-      "required_pages_per_second_upper_estimate": 0.004006553397533954
+      "required_token_positions_per_second": 19323.088757619076,
+      "required_pages_per_second_upper_estimate": 0.004024489804093467
     },
     {
       "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": 28855.453988456982,
-      "required_pages_per_second_upper_estimate": 0.0017298404068808967
+      "required_token_positions_per_second": 28984.633136428616,
+      "required_pages_per_second_upper_estimate": 0.0017375844995566581
     },
     {
       "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": 28855.453988456982,
-      "required_pages_per_second_upper_estimate": 0.002243384625873937
+      "required_token_positions_per_second": 28984.633136428616,
+      "required_pages_per_second_upper_estimate": 0.0022534277364297086
     },
     {
       "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": 28855.453988456982,
-      "required_pages_per_second_upper_estimate": 0.3701267811913262
+      "required_token_positions_per_second": 28984.633136428616,
+      "required_pages_per_second_upper_estimate": 0.37178375259974367
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.33756897937500696,
-    "producer_p95_seconds": 0.24165862599875254,
-    "gpu_update_seconds": 0.9581550860735029,
-    "producer_batches_per_second": 2.962357506461206,
-    "gpu_batches_per_second": 1.0436723809482416,
-    "producer_headroom": 2.8383979115838245,
+    "producer_batch_seconds": 0.08076144157018916,
+    "producer_p95_seconds": 0.2235941129983985,
+    "gpu_update_seconds": 0.9538847661056402,
+    "producer_batches_per_second": 12.382146486710585,
+    "gpu_batches_per_second": 1.048344659159021,
+    "producer_headroom": 11.811140905361702,
     "sustainable": true,
-    "buffer_seconds": 60.36377042263068,
+    "buffer_seconds": 60.094740264655336,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 438122.9119123068,
+    "generation_bytes_per_second": 456073.34529352654,
     "upload_bytes_per_second": 1886866.1050289606,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.0109779169741,
+    "estimated_full_shards_per_hour": 97.86272305588099,
     "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-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -162,3 +162,4 @@
 {"step": 661200, "loss": 0.5655838847160339, "lr": 0.0001, "grad_norm": 0.18529066629640334, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96782162489, "tokens_per_second": 154032.61333851627, "data_wait_s": 8.934999641496688e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 661300, "loss": 0.501309871673584, "lr": 0.0001, "grad_norm": 0.1910924292280799, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96796881185, "tokens_per_second": 153321.34276279656, "data_wait_s": 1.4727000234415755e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 661400, "loss": 0.5986628532409668, "lr": 0.0001, "grad_norm": 0.20128587924484945, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96811615482, "tokens_per_second": 154072.87834906369, "data_wait_s": 8.910001270123757e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 661500, "loss": 0.5121641159057617, "lr": 0.0001, "grad_norm": 0.19666728712620124, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96826378579, "tokens_per_second": 153557.48604135014, "data_wait_s": 9.351999324280769e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d76d25d815453d6fcc3f0f1c26c3d041507511839e62dd3c051c3270aa82733f
-size 349866
+oid sha256:85d2f3dea8071fc4214805e362cf0f8f43536f7def58cda53b818c6be617011a
+size 351984