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": 76525.33320633018,
-      "required_pages_per_second_upper_estimate": 0.0052942001144504
+      "required_token_positions_per_second": 76488.96186925404,
+      "required_pages_per_second_upper_estimate": 0.005291683860958344
     },
     {
       "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": 9565.666650791272,
-      "required_pages_per_second_upper_estimate": 0.004004944872851871
+      "required_token_positions_per_second": 9561.120233656755,
+      "required_pages_per_second_upper_estimate": 0.0040030413829376344
     },
     {
       "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": 19131.333301582545,
-      "required_pages_per_second_upper_estimate": 0.003984552199532498
+      "required_token_positions_per_second": 19122.24046731351,
+      "required_pages_per_second_upper_estimate": 0.0039826584019484225
     },
     {
       "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": 28696.999952373815,
-      "required_pages_per_second_upper_estimate": 0.0017203413293630147
+      "required_token_positions_per_second": 28683.360700970265,
+      "required_pages_per_second_upper_estimate": 0.001719523677067303
     },
     {
       "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": 28696.999952373815,
-      "required_pages_per_second_upper_estimate": 0.002231065521534118
+      "required_token_positions_per_second": 28683.360700970265,
+      "required_pages_per_second_upper_estimate": 0.0022300051297302185
     },
     {
       "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": 28696.999952373815,
-      "required_pages_per_second_upper_estimate": 0.3680943029511399
+      "required_token_positions_per_second": 28683.360700970265,
+      "required_pages_per_second_upper_estimate": 0.3679193532788223
     }
   ],
   "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": 0.09464874800780265,
-    "producer_p95_seconds": 0.14375834999987092,
-    "gpu_update_seconds": 0.96344565793934,
-    "producer_batches_per_second": 10.565380113824242,
-    "gpu_batches_per_second": 1.0379412598514834,
-    "producer_headroom": 10.179169595142616,
+    "producer_batch_seconds": 0.04327865742970438,
+    "producer_p95_seconds": 0.13350533800007724,
+    "gpu_update_seconds": 0.9639037868761577,
+    "producer_batches_per_second": 23.106077207322247,
+    "gpu_batches_per_second": 1.037447942020047,
+    "producer_headroom": 22.27203531999079,
     "sustainable": true,
-    "buffer_seconds": 60.69707645017842,
+    "buffer_seconds": 60.725938573197936,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 63,
+    "upload_pages_per_commit": 67,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 471703.2481933078,
-    "upload_bytes_per_second": 5495978.476499954,
+    "generation_bytes_per_second": 469148.47237436456,
+    "upload_bytes_per_second": 5635516.838172999,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.21653637265611,
+    "estimated_full_shards_per_hour": 100.66834095404818,
     "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": [
@@ -220,6 +220,12 @@
       "bytes": 8072363,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 0.9177546119999533,
+      "bytes": 6210206,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790911926322536473-f9bff804/train.jsonl

@@ -9,3 +9,4 @@
 {"step": 484100, "loss": 0.5748189091682434, "lr": 0.0003, "grad_norm": 0.15505583909122223, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70855537479, "tokens_per_second": 152773.91011407136, "data_wait_s": 7.912999990367098e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 484200, "loss": 0.6031655073165894, "lr": 0.0003, "grad_norm": 0.16529147406275682, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70870343274, "tokens_per_second": 148905.85149239417, "data_wait_s": 2.171400001316215e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 484300, "loss": 0.5535135865211487, "lr": 0.0003, "grad_norm": 0.1321929690668786, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70885166133, "tokens_per_second": 153684.04961715933, "data_wait_s": 1.0803999884956283e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 484400, "loss": 0.6193707585334778, "lr": 0.0003, "grad_norm": 0.164754628098268, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70899882212, "tokens_per_second": 152979.286054062, "data_wait_s": 1.561599992783158e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790911926.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.444.0.1790911926322536473-f9bff804

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0644dfcfdb61731094122c242a6eb35f35855fce50d976ed20b831057614d526
-size 23463
+oid sha256:761ec265fbf69f91d53ce3c0bf884e47b70691bd15c1f972a08e758a7e587025
+size 25581