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": 77260.49040310859,
-      "required_pages_per_second_upper_estimate": 0.005345059995123241
+      "required_token_positions_per_second": 77291.63280744193,
+      "required_pages_per_second_upper_estimate": 0.005347214498915355
     },
     {
       "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": 9657.561300388574,
-      "required_pages_per_second_upper_estimate": 0.004043419243659764
+      "required_token_positions_per_second": 9661.454100930241,
+      "required_pages_per_second_upper_estimate": 0.00404504907795564
     },
     {
       "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": 19315.122600777147,
-      "required_pages_per_second_upper_estimate": 0.004022830663704977
+      "required_token_positions_per_second": 19322.908201860482,
+      "required_pages_per_second_upper_estimate": 0.0040244521990905275
     },
     {
       "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": 28972.683901165718,
-      "required_pages_per_second_upper_estimate": 0.001736868161148073
+      "required_token_positions_per_second": 28984.362302790723,
+      "required_pages_per_second_upper_estimate": 0.0017375682634936053
     },
     {
       "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": 28972.683901165718,
-      "required_pages_per_second_upper_estimate": 0.0022524987359471495
+      "required_token_positions_per_second": 28984.362302790723,
+      "required_pages_per_second_upper_estimate": 0.0022534066803056338
     },
     {
       "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": 28972.683901165718,
-      "required_pages_per_second_upper_estimate": 0.3716304806398804
+      "required_token_positions_per_second": 28984.362302790723,
+      "required_pages_per_second_upper_estimate": 0.37178027863663526
     }
   ],
   "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": 0.14804104100000615,
-    "producer_p95_seconds": 0.15726751400006833,
-    "gpu_update_seconds": 0.9542781778283088,
-    "producer_batches_per_second": 6.754883600149491,
-    "gpu_batches_per_second": 1.0479124674900795,
-    "producer_headroom": 6.446038013392983,
+    "producer_batch_seconds": 0.08003096183593428,
+    "producer_p95_seconds": 0.15656870099974185,
+    "gpu_update_seconds": 0.9538936793285235,
+    "producer_batches_per_second": 12.495164084745452,
+    "gpu_batches_per_second": 1.0483348633821876,
+    "producer_headroom": 11.919058042611463,
     "sustainable": true,
-    "buffer_seconds": 60.11952520318346,
+    "buffer_seconds": 60.09530179769698,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 405190.5185956452,
-    "upload_bytes_per_second": 546928.4410035238,
+    "generation_bytes_per_second": 400167.58987785963,
+    "upload_bytes_per_second": 699447.9924095691,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.94445293809908,
+    "estimated_full_shards_per_hour": 85.86664936305849,
     "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": [
@@ -202,6 +202,12 @@
       "bytes": 6259058,
       "pages": 42,
       "shards": 2
+    },
+    {
+      "seconds": 1.4518450940004186,
+      "bytes": 6815038,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -36,3 +36,4 @@
 {"step": 648600, "loss": 0.5634985566139221, "lr": 0.0001, "grad_norm": 0.17079177724091404, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94938416805, "tokens_per_second": 152800.09515960494, "data_wait_s": 1.3675999980478082e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 648700, "loss": 0.546423614025116, "lr": 0.0001, "grad_norm": 0.1903061926083244, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94953087129, "tokens_per_second": 152879.50500266466, "data_wait_s": 1.6120000054797856e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 648800, "loss": 0.5665684938430786, "lr": 0.0001, "grad_norm": 0.17694188959309323, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94967773194, "tokens_per_second": 152769.83153522035, "data_wait_s": 1.3683000361197628e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 648900, "loss": 0.5581087470054626, "lr": 0.0001, "grad_norm": 0.17167866499828188, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94982472685, "tokens_per_second": 153190.41151221056, "data_wait_s": 1.2909999895782676e-05, "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 sha256:3dd8f5b118947d018f8162a7ea7d06ac937b40689c3871dd34e9e30c11914812
-size 80649
+oid sha256:31b849eb59d941d249bf66226b245d28b4c47f26b3c5fd6778be6df63b17d10f
+size 82767