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": 77318.67340671357,
-      "required_pages_per_second_upper_estimate": 0.005349085230315801
+      "required_token_positions_per_second": 77066.01917415895,
+      "required_pages_per_second_upper_estimate": 0.005331606024269085
     },
     {
       "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": 9664.834175839196,
-      "required_pages_per_second_upper_estimate": 0.00404646424473603
+      "required_token_positions_per_second": 9633.252396769869,
+      "required_pages_per_second_upper_estimate": 0.004033241613342244
     },
     {
       "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": 19329.668351678392,
-      "required_pages_per_second_upper_estimate": 0.00402586016002046
+      "required_token_positions_per_second": 19266.504793539738,
+      "required_pages_per_second_upper_estimate": 0.004012704856595222
     },
     {
       "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": 28994.50252751759,
-      "required_pages_per_second_upper_estimate": 0.0017381761544827544
+      "required_token_positions_per_second": 28899.757190309607,
+      "required_pages_per_second_upper_estimate": 0.001732496316184889
     },
     {
       "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": 28994.50252751759,
-      "required_pages_per_second_upper_estimate": 0.0022541950381760128
+      "required_token_positions_per_second": 28899.757190309607,
+      "required_pages_per_second_upper_estimate": 0.002246829004948788
     },
     {
       "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": 28994.50252751759,
-      "required_pages_per_second_upper_estimate": 0.3719103465517065
+      "required_token_positions_per_second": 28899.757190309607,
+      "required_pages_per_second_upper_estimate": 0.3706950550956197
     }
   ],
   "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": 0.0627136116718674,
-    "producer_p95_seconds": 0.14201421899997513,
-    "gpu_update_seconds": 0.9535600748369566,
-    "producer_batches_per_second": 15.94550167565279,
-    "gpu_batches_per_second": 1.0487016249825518,
-    "producer_headroom": 15.204993771148292,
+    "producer_batch_seconds": 0.22568241228902508,
+    "producer_p95_seconds": 0.1525195469994287,
+    "gpu_update_seconds": 0.9566862385013624,
+    "producer_batches_per_second": 4.431005455220535,
+    "gpu_batches_per_second": 1.0452747826356195,
+    "producer_headroom": 4.23908194173395,
     "sustainable": true,
-    "buffer_seconds": 60.074284714728265,
+    "buffer_seconds": 60.27123302558583,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 364861.80810673145,
-    "upload_bytes_per_second": 5668609.309602238,
+    "generation_bytes_per_second": 349094.54780491634,
+    "upload_bytes_per_second": 5667059.701029967,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.29085047151048,
+    "estimated_full_shards_per_hour": 74.90756345377558,
     "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": [
@@ -250,6 +250,12 @@
       "bytes": 2279800,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.0028891179999846,
+      "bytes": 5661960,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -18,3 +18,4 @@
 {"step": 486800, "loss": 0.5181353688240051, "lr": 0.0003, "grad_norm": 0.17681517005659952, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71251312946, "tokens_per_second": 152511.9095738598, "data_wait_s": 1.0664000001270324e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 486900, "loss": 0.5777289271354675, "lr": 0.0003, "grad_norm": 0.16178990640805685, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71265952505, "tokens_per_second": 153478.87240364606, "data_wait_s": 8.304999937536195e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 487000, "loss": 0.550245463848114, "lr": 0.0003, "grad_norm": 0.15350373814620724, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71280588623, "tokens_per_second": 152684.9476076727, "data_wait_s": 8.385999535676092e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 487100, "loss": 0.5828955769538879, "lr": 0.0003, "grad_norm": 0.17236955307801755, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71295318846, "tokens_per_second": 153595.28166588835, "data_wait_s": 1.0404000022390392e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8a662d89fb9cf2ac00914dbc4c0ecfaf7997c652bdc1f51a36e8e990727f5edb
-size 42525
+oid sha256:cceb4fa13e6f1691633ff0b7ceac6b164b489561d079e37ef4fc71344c7eff68
+size 44643