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": 75335.94636392311,
-      "required_pages_per_second_upper_estimate": 0.006529928893614429
+      "required_token_positions_per_second": 75355.68685040352,
+      "required_pages_per_second_upper_estimate": 0.006531639948950745
     },
     {
       "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": 9416.993295490389,
-      "required_pages_per_second_upper_estimate": 0.003756579422167859
+      "required_token_positions_per_second": 9419.46085630044,
+      "required_pages_per_second_upper_estimate": 0.00375756376906831
     },
     {
       "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": 18833.986590980778,
-      "required_pages_per_second_upper_estimate": 0.0043569113309804
+      "required_token_positions_per_second": 18838.92171260088,
+      "required_pages_per_second_upper_estimate": 0.004358052984512037
     },
     {
       "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": 28250.979886471167,
-      "required_pages_per_second_upper_estimate": 0.001720336453649786
+      "required_token_positions_per_second": 28258.38256890132,
+      "required_pages_per_second_upper_estimate": 0.0017207872381709093
     },
     {
       "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": 28250.979886471167,
-      "required_pages_per_second_upper_estimate": 0.002174838635378137
+      "required_token_positions_per_second": 28258.38256890132,
+      "required_pages_per_second_upper_estimate": 0.0021754085143635467
     },
     {
       "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": 28250.979886471167,
-      "required_pages_per_second_upper_estimate": 0.3507826202425117
+      "required_token_positions_per_second": 28258.38256890132,
+      "required_pages_per_second_upper_estimate": 0.35087453678310987
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.16328131734337603,
-    "producer_p95_seconds": 0.4133406179971644,
-    "gpu_update_seconds": 0.978656319572125,
-    "producer_batches_per_second": 6.124399387941169,
-    "gpu_batches_per_second": 1.0218091683474815,
-    "producer_headroom": 5.993682164592279,
+    "producer_batch_seconds": 0.45454338560108454,
+    "producer_p95_seconds": 0.41785758700279985,
+    "gpu_update_seconds": 0.97839994672685,
+    "producer_batches_per_second": 2.20001001373633,
+    "gpu_batches_per_second": 1.0220769158312109,
+    "producer_headroom": 2.152489680238162,
     "sustainable": true,
-    "buffer_seconds": 60.67669181347175,
+    "buffer_seconds": 60.660796697064704,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 326728.9085583137,
+    "generation_bytes_per_second": 302881.05466962303,
     "upload_bytes_per_second": 5974885.961383253,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.10841791688974,
+    "estimated_full_shards_per_hour": 64.99122362200278,
     "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-v1/1790485060735632541-7579de36/train.jsonl

@@ -342,3 +342,4 @@
 {"step": 119200, "loss": 0.5859232051298022, "lr": 0.0003, "grad_norm": 0.13938095464892203, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17443251462, "tokens_per_second": 148488.68158974353, "data_wait_s": 1.6275000234600157e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 119300, "loss": 0.6423541735857725, "lr": 0.0003, "grad_norm": 0.13440198642835463, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17457970719, "tokens_per_second": 149450.31556604808, "data_wait_s": 1.8403996364213526e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 119400, "loss": 0.6104825809597969, "lr": 0.0003, "grad_norm": 0.16794767334880456, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17472747252, "tokens_per_second": 149110.36216746346, "data_wait_s": 1.1146999895572662e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 119500, "loss": 0.5576545484364033, "lr": 0.0003, "grad_norm": 0.12738941614743302, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17487473060, "tokens_per_second": 149518.90413774241, "data_wait_s": 1.3624994608107954e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:444363139c9ac6cb291f0eb1dad205bd9443c10db98177c266ae43b1718d3317
-size 733455
+oid sha256:d5dd603c7b9c01d3566c9c174a7688f038138f1e7ca15d4d6dedaf9e7f0b985f
+size 735573