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": 75581.59962706485,
-      "required_pages_per_second_upper_estimate": 0.00522891043567323
+      "required_token_positions_per_second": 76401.99901579303,
+      "required_pages_per_second_upper_estimate": 0.005285667568974291
     },
     {
       "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": 9447.699953383106,
-      "required_pages_per_second_upper_estimate": 0.003955554680071839
+      "required_token_positions_per_second": 9550.249876974129,
+      "required_pages_per_second_upper_estimate": 0.003998490191593479
     },
     {
       "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": 18895.39990676621,
-      "required_pages_per_second_upper_estimate": 0.003935413495374287
+      "required_token_positions_per_second": 19100.499753948257,
+      "required_pages_per_second_upper_estimate": 0.003978130384695608
     },
     {
       "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": 28343.099860149316,
-      "required_pages_per_second_upper_estimate": 0.0016991255592083082
+      "required_token_positions_per_second": 28650.749630922386,
+      "required_pages_per_second_upper_estimate": 0.0017175686932121535
     },
     {
       "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": 28343.099860149316,
-      "required_pages_per_second_upper_estimate": 0.0022035513460056575
+      "required_token_positions_per_second": 28650.749630922386,
+      "required_pages_per_second_upper_estimate": 0.0022274697624749336
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28343.099860149316,
-      "required_pages_per_second_upper_estimate": 0.36355485255639763
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28650.749630922386,
+      "required_pages_per_second_upper_estimate": 0.36750105348728707
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "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,
+    "producer_batch_seconds": 0.10267891665631623,
+    "producer_p95_seconds": 0.2835509659998934,
+    "gpu_update_seconds": 0.9650009286374787,
+    "producer_batches_per_second": 9.739097689812699,
+    "gpu_batches_per_second": 1.036268432831394,
+    "producer_headroom": 9.398238314760377,
     "sustainable": true,
-    "buffer_seconds": 60.479482077052154,
+    "buffer_seconds": 60.795058504161155,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 412852.80957922246,
-    "upload_bytes_per_second": 5966798.679373037,
+    "generation_bytes_per_second": 427142.4917304415,
+    "upload_bytes_per_second": 5887256.641678577,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.58860221416955,
+    "estimated_full_shards_per_hour": 91.6548353570455,
     "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": 1.0216239910005243,
-      "bytes": 7839111,
-      "pages": 37,
-      "shards": 4
-    },
     {
       "seconds": 1.0000162349970196,
       "bytes": 885768,
@@ -562,6 +556,12 @@
       "bytes": 13228107,
       "pages": 34,
       "shards": 4
+    },
+    {
+      "seconds": 0.8756032579985913,
+      "bytes": 1232264,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -270,3 +270,4 @@
 {"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}
+{"step": 424400, "loss": 0.6360502243041992, "lr": 0.0003, "grad_norm": 0.16612847388856186, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62120496160, "tokens_per_second": 150041.53417072596, "data_wait_s": 1.3700999261345714e-05, "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:dbdc8f1d9695d39faa06a50d84e6e9e0166b1814cb175216e131a42db6084b12
-size 580176
+oid sha256:1d4a4fb8268c1394715e75f25090b70a583143396400d6768fce361f45582bc1
+size 582294