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": 77168.20254439136,
-      "required_pages_per_second_upper_estimate": 0.005338675306919848
+      "required_token_positions_per_second": 77625.41353493286,
+      "required_pages_per_second_upper_estimate": 0.005370306224121167
     },
     {
       "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": 9646.02531804892,
-      "required_pages_per_second_upper_estimate": 0.004038589368752855
+      "required_token_positions_per_second": 9703.176691866607,
+      "required_pages_per_second_upper_estimate": 0.004062517455513923
     },
     {
       "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": 19292.05063609784,
-      "required_pages_per_second_upper_estimate": 0.004018025381910902
+      "required_token_positions_per_second": 19406.353383733214,
+      "required_pages_per_second_upper_estimate": 0.004041831629877188
     },
     {
       "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": 28938.07595414676,
-      "required_pages_per_second_upper_estimate": 0.0017347934675675618
+      "required_token_positions_per_second": 29109.53007559982,
+      "required_pages_per_second_upper_estimate": 0.001745071880353388
     },
     {
       "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": 28938.07595414676,
-      "required_pages_per_second_upper_estimate": 0.002249808120290697
+      "required_token_positions_per_second": 29109.53007559982,
+      "required_pages_per_second_upper_estimate": 0.002263137924086693
     },
     {
       "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": 28938.07595414676,
-      "required_pages_per_second_upper_estimate": 0.37118656705463965
+      "required_token_positions_per_second": 29109.53007559982,
+      "required_pages_per_second_upper_estimate": 0.3733857964315468
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.7857319563750877,
-    "producer_p95_seconds": 0.2605938510005217,
-    "gpu_update_seconds": 0.9554194288455485,
-    "producer_batches_per_second": 1.2726986498212711,
-    "gpu_batches_per_second": 1.0466607332952387,
-    "producer_headroom": 1.2159610171047395,
+    "producer_batch_seconds": 0.11694373137495973,
+    "producer_p95_seconds": 0.23704230799921788,
+    "gpu_update_seconds": 0.9497920415821173,
+    "producer_batches_per_second": 8.551121024124619,
+    "gpu_batches_per_second": 1.0528620542389981,
+    "producer_headroom": 8.121786695319088,
     "sustainable": true,
-    "buffer_seconds": 60.191424017269554,
+    "buffer_seconds": 60.786690661255506,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 272028.8727644594,
-    "upload_bytes_per_second": 5185689.43209054,
+    "generation_bytes_per_second": 271920.24606028467,
+    "upload_bytes_per_second": 5244487.848015167,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.371063587191934,
+    "estimated_full_shards_per_hour": 58.34775482517629,
     "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": [
-    {
-      "seconds": 1.6983987419989717,
-      "bytes": 12559131,
-      "pages": 36,
-      "shards": 4
-    },
     {
       "seconds": 1.1707839329992566,
       "bytes": 6060902,
@@ -562,6 +556,12 @@
       "bytes": 6045695,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.710491572997853,
+      "bytes": 17528313,
+      "pages": 40,
+      "shards": 5
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -164,3 +164,4 @@
 {"step": 901400, "loss": 0.5353484749794006, "lr": 3e-05, "grad_norm": 0.3146727884420038, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131939470110, "tokens_per_second": 153312.7049090487, "data_wait_s": 9.49600143940188e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 901500, "loss": 0.5521841049194336, "lr": 3e-05, "grad_norm": 0.3143202209296977, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131953978946, "tokens_per_second": 150168.55357323462, "data_wait_s": 1.8426999304210767e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 901600, "loss": 0.4958744943141937, "lr": 3e-05, "grad_norm": 0.3201443174944292, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131968518105, "tokens_per_second": 152118.64782818613, "data_wait_s": 8.743001671973616e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 901700, "loss": 0.5292997360229492, "lr": 3e-05, "grad_norm": 0.31630968411192373, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131983219989, "tokens_per_second": 153078.70515092195, "data_wait_s": 1.0854000720428303e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ed80ae67fdd7508c98b5dd691c10976e34bf74c1db2be6511a8569ef6cbd5434
-size 354102
+oid sha256:4662d8c96364ca452a048ce9f6ccd89509f7577ed1d677b4a64bffcf8d851e17
+size 356220