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": 77289.02864536893,
-      "required_pages_per_second_upper_estimate": 0.0053470343369406564
+      "required_token_positions_per_second": 77173.6762567645,
+      "required_pages_per_second_upper_estimate": 0.005339053991042585
     },
     {
       "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": 9661.128580671117,
-      "required_pages_per_second_upper_estimate": 0.004044912789420781
+      "required_token_positions_per_second": 9646.709532095563,
+      "required_pages_per_second_upper_estimate": 0.0040388758348861705
     },
     {
       "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": 19322.257161342233,
-      "required_pages_per_second_upper_estimate": 0.00402431660451967
+      "required_token_positions_per_second": 19293.419064191126,
+      "required_pages_per_second_upper_estimate": 0.004018310389394858
     },
     {
       "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": 28983.38574201335,
-      "required_pages_per_second_upper_estimate": 0.0017375097201661237
+      "required_token_positions_per_second": 28940.12859628669,
+      "required_pages_per_second_upper_estimate": 0.0017349165203296476
     },
     {
       "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": 28983.38574201335,
-      "required_pages_per_second_upper_estimate": 0.0022533307570006297
+      "required_token_positions_per_second": 28940.12859628669,
+      "required_pages_per_second_upper_estimate": 0.002249967704188458
     },
     {
       "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": 28983.38574201335,
-      "required_pages_per_second_upper_estimate": 0.37176775236353243
+      "required_token_positions_per_second": 28940.12859628669,
+      "required_pages_per_second_upper_estimate": 0.3712128961440552
     }
   ],
   "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": 0.08486304620319629,
-    "producer_p95_seconds": 0.25102540499938186,
-    "gpu_update_seconds": 0.9539258196437134,
-    "producer_batches_per_second": 11.783692016023057,
-    "gpu_batches_per_second": 1.0482995421735153,
-    "producer_headroom": 11.240768064813876,
+    "producer_batch_seconds": 0.39438611727348416,
+    "producer_p95_seconds": 0.2660703999972611,
+    "gpu_update_seconds": 0.9553516636255555,
+    "producer_batches_per_second": 2.5355862090514645,
+    "gpu_batches_per_second": 1.046734975270786,
+    "producer_headroom": 2.422376503083332,
     "sustainable": true,
-    "buffer_seconds": 60.09732663755394,
+    "buffer_seconds": 60.18715480840999,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 356324.6006961768,
+    "generation_bytes_per_second": 315445.8416899396,
     "upload_bytes_per_second": 5404764.988606308,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.4589644972227,
+    "estimated_full_shards_per_hour": 67.68733442329065,
     "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-v1/1790989041640695387-36510d53/train.jsonl

@@ -250,3 +250,4 @@
 {"step": 582100, "loss": 0.613437831401825, "lr": 0.0003, "grad_norm": 0.18855394660947497, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85202590969, "tokens_per_second": 153180.8067054749, "data_wait_s": 1.205899752676487e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 582200, "loss": 0.6435660719871521, "lr": 0.0003, "grad_norm": 0.18179106690589025, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85217344932, "tokens_per_second": 153247.3753550792, "data_wait_s": 7.557002390967682e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 582300, "loss": 0.5519033074378967, "lr": 0.0003, "grad_norm": 0.17627105548430186, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85231989464, "tokens_per_second": 153438.6389978105, "data_wait_s": 1.1235999409109354e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 582400, "loss": 0.587166965007782, "lr": 0.0003, "grad_norm": 0.16023193403620564, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85246743268, "tokens_per_second": 152593.23711836973, "data_wait_s": 1.982600224437192e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b14c1c8926fd5fbbbad0a8cd2d0cab160a4b1fe6296705175e7bd32737220b8c
-size 537816
+oid sha256:44dd195a821c6deaf17f6b62051299127034be324b773dd5212812dfda88287b
+size 539934