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": 75822.40961744744,
-      "required_pages_per_second_upper_estimate": 0.005245570229564058
+      "required_token_positions_per_second": 75581.59962706485,
+      "required_pages_per_second_upper_estimate": 0.00522891043567323
     },
     {
       "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": 9477.80120218093,
-      "required_pages_per_second_upper_estimate": 0.003968157444357935
+      "required_token_positions_per_second": 9447.699953383106,
+      "required_pages_per_second_upper_estimate": 0.003955554680071839
     },
     {
       "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": 18955.60240436186,
-      "required_pages_per_second_upper_estimate": 0.003947952087976834
+      "required_token_positions_per_second": 18895.39990676621,
+      "required_pages_per_second_upper_estimate": 0.003935413495374287
     },
     {
       "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": 28433.403606542794,
-      "required_pages_per_second_upper_estimate": 0.0017045391309187603
+      "required_token_positions_per_second": 28343.099860149316,
+      "required_pages_per_second_upper_estimate": 0.0016991255592083082
     },
     {
       "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": 28433.403606542794,
-      "required_pages_per_second_upper_estimate": 0.002210572065083548
+      "required_token_positions_per_second": 28343.099860149316,
+      "required_pages_per_second_upper_estimate": 0.0022035513460056575
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28433.403606542794,
-      "required_pages_per_second_upper_estimate": 0.36471317205452464
+      "required_token_positions_per_second": 28343.099860149316,
+      "required_pages_per_second_upper_estimate": 0.36355485255639763
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.38157920347669005,
-    "producer_p95_seconds": 0.28838687199822743,
-    "gpu_update_seconds": 0.9723774326348301,
-    "producer_batches_per_second": 2.6206878962183486,
-    "gpu_batches_per_second": 1.028407248500535,
-    "producer_headroom": 2.548297768261972,
+    "producer_batch_seconds": 0.12459814880469366,
+    "producer_p95_seconds": 0.27847433499846375,
+    "gpu_update_seconds": 0.9754755173718089,
+    "producer_batches_per_second": 8.02580142316151,
+    "gpu_batches_per_second": 1.025141053969521,
+    "producer_headroom": 7.828972795581874,
     "sustainable": true,
-    "buffer_seconds": 60.28740082335947,
+    "buffer_seconds": 60.479482077052154,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 429427.70277391275,
-    "upload_bytes_per_second": 5894493.727171952,
+    "generation_bytes_per_second": 412852.80957922246,
+    "upload_bytes_per_second": 5966798.679373037,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.14518844998395,
+    "estimated_full_shards_per_hour": 88.58860221416955,
     "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": 1.206743112001277,
-      "bytes": 6257240,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0216239910005243,
       "bytes": 7839111,
@@ -562,6 +556,12 @@
       "bytes": 10259217,
       "pages": 40,
       "shards": 4
+    },
+    {
+      "seconds": 1.5030511410004692,
+      "bytes": 13228107,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -269,3 +269,4 @@
 {"step": 424000, "loss": 0.6058544516563416, "lr": 0.0003, "grad_norm": 0.17613054731147765, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62062117848, "tokens_per_second": 153140.9930948525, "data_wait_s": 1.4223001926438883e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 424100, "loss": 0.5639068484306335, "lr": 0.0003, "grad_norm": 0.14342668742039902, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62076902751, "tokens_per_second": 152433.02420459464, "data_wait_s": 1.112700192607008e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 424200, "loss": 0.6424958109855652, "lr": 0.0003, "grad_norm": 0.14314316147947068, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62091558430, "tokens_per_second": 151029.45185137034, "data_wait_s": 1.0495001333765686e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 424300, "loss": 0.5703352689743042, "lr": 0.0003, "grad_norm": 0.16307918388913314, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62106010392, "tokens_per_second": 150076.1714774224, "data_wait_s": 7.349000952672213e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9d1b117d24e3249857e9d1b2250158c728150d632b7b54bc2a49c73ccf6865ad
-size 578058
+oid sha256:dbdc8f1d9695d39faa06a50d84e6e9e0166b1814cb175216e131a42db6084b12
+size 580176