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": 76618.30310808889,
-      "required_pages_per_second_upper_estimate": 0.0053006319879739605
+      "required_token_positions_per_second": 76110.88812693022,
+      "required_pages_per_second_upper_estimate": 0.0052655278422124386
     },
     {
       "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": 9577.287888511111,
-      "required_pages_per_second_upper_estimate": 0.004009810442406128
+      "required_token_positions_per_second": 9513.861015866278,
+      "required_pages_per_second_upper_estimate": 0.003983254935333452
     },
     {
       "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": 19154.575777022223,
-      "required_pages_per_second_upper_estimate": 0.0039893929942212865
+      "required_token_positions_per_second": 19027.722031732555,
+      "required_pages_per_second_upper_estimate": 0.003962972704435678
     },
     {
       "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": 28731.863665533336,
-      "required_pages_per_second_upper_estimate": 0.0017224313557331204
+      "required_token_positions_per_second": 28541.583047598833,
+      "required_pages_per_second_upper_estimate": 0.0017110243232296278
     },
     {
       "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": 28731.863665533336,
-      "required_pages_per_second_upper_estimate": 0.0022337760218830026
+      "required_token_positions_per_second": 28541.583047598833,
+      "required_pages_per_second_upper_estimate": 0.0022189825407945814
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28731.863665533336,
-      "required_pages_per_second_upper_estimate": 0.3685414972298115
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28541.583047598833,
+      "required_pages_per_second_upper_estimate": 0.3661007817703574
     }
   ],
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.18720627861750927,
-    "producer_p95_seconds": 0.3657566580004641,
-    "gpu_update_seconds": 0.9622765972249292,
-    "producer_batches_per_second": 5.34170118323409,
-    "gpu_batches_per_second": 1.0392022448471259,
-    "producer_headroom": 5.1401940379948785,
+    "producer_batch_seconds": 0.34408107491418605,
+    "producer_p95_seconds": 0.3611759070045082,
+    "gpu_update_seconds": 0.968691889090083,
+    "producer_batches_per_second": 2.9062917809397115,
+    "gpu_batches_per_second": 1.032319988700768,
+    "producer_headroom": 2.8153012755254707,
     "sustainable": true,
-    "buffer_seconds": 60.62342562517054,
+    "buffer_seconds": 60.05889712358515,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 400717.772618566,
-    "upload_bytes_per_second": 5028107.140468847,
+    "generation_bytes_per_second": 380985.1779224774,
+    "upload_bytes_per_second": 5033920.350783255,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.98470577161537,
+    "estimated_full_shards_per_hour": 81.75055030112973,
     "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": 1.0187819040002069,
-      "bytes": 1294212,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.2491005060001044,
       "bytes": 11541986,
@@ -562,6 +556,12 @@
       "bytes": 11833830,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.1506007540010614,
+      "bytes": 2440099,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -363,3 +363,4 @@
 {"step": 476500, "loss": 0.604406476020813, "lr": 0.0003, "grad_norm": 0.16532843781642648, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69744435689, "tokens_per_second": 152133.76748892048, "data_wait_s": 9.391995263285935e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 476600, "loss": 0.5556173920631409, "lr": 0.0003, "grad_norm": 0.14372095280793437, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69759169013, "tokens_per_second": 152997.47451210505, "data_wait_s": 9.937000868376344e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 476700, "loss": 0.6041773557662964, "lr": 0.0003, "grad_norm": 0.18339025125092456, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69773730765, "tokens_per_second": 151307.86727909005, "data_wait_s": 1.4555997040588409e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 476800, "loss": 0.6477830410003662, "lr": 0.0003, "grad_norm": 0.17604146039098187, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69788381008, "tokens_per_second": 151402.06029228034, "data_wait_s": 9.354000212624669e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:04dd5412c80947c8fc9ed41d9859f1a14cbb06ac06db25fa0500b488e639168e
-size 778716
+oid sha256:09533c05f38280ef7f01e9be4ea55698f778ba8cba3f5550394d3c8d57c32680
+size 780834