Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75731.85763252276,
-      "required_pages_per_second_upper_estimate": 0.00656424547894871
+      "required_token_positions_per_second": 74776.44346701661,
+      "required_pages_per_second_upper_estimate": 0.006481432600557701
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9466.482204065345,
-      "required_pages_per_second_upper_estimate": 0.00377632128772353
+      "required_token_positions_per_second": 9347.055433377076,
+      "required_pages_per_second_upper_estimate": 0.0037286801633066363
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18932.96440813069,
-      "required_pages_per_second_upper_estimate": 0.004379808107027945
+      "required_token_positions_per_second": 18694.110866754152,
+      "required_pages_per_second_upper_estimate": 0.004324553543909238
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28399.446612196036,
-      "required_pages_per_second_upper_estimate": 0.0017293772983017206
+      "required_token_positions_per_second": 28041.16630013123,
+      "required_pages_per_second_upper_estimate": 0.0017075599070484964
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28399.446612196036,
-      "required_pages_per_second_upper_estimate": 0.002186268014906638
+      "required_token_positions_per_second": 28041.16630013123,
+      "required_pages_per_second_upper_estimate": 0.0021586866047005084
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28399.446612196036,
-      "required_pages_per_second_upper_estimate": 0.35262608008984736
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28041.16630013123,
+      "required_pages_per_second_upper_estimate": 0.3481774377010719
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.1143375932577726,
-    "producer_p95_seconds": 0.2546618079995824,
-    "gpu_update_seconds": 0.9735400966625409,
-    "producer_batches_per_second": 8.74602981843,
-    "gpu_batches_per_second": 1.0271790586008405,
-    "producer_headroom": 8.514610714847807,
+    "producer_batch_seconds": 0.057545325234315214,
+    "producer_p95_seconds": 0.19222766600069008,
+    "gpu_update_seconds": 0.9859789605067394,
+    "producer_batches_per_second": 17.377606190045196,
+    "gpu_batches_per_second": 1.014220424628589,
+    "producer_headroom": 17.13395408735624,
     "sustainable": true,
-    "buffer_seconds": 60.35948599307754,
+    "buffer_seconds": 60.1447165909111,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 219235.59043132063,
+    "generation_bytes_per_second": 233500.85525451138,
     "upload_bytes_per_second": 6474041.960848427,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 47.04285416321482,
+    "estimated_full_shards_per_hour": 50.10384791590219,
     "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": [

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -109,3 +109,4 @@
 {"step": 10900, "loss": 0.6950379349291325, "lr": 0.0003, "grad_norm": 0.15290236586700576, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1594471529, "tokens_per_second": 148803.07798327558, "data_wait_s": 1.2014999811071903e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 11000, "loss": 0.722653703764081, "lr": 0.0003, "grad_norm": 0.14722297485559593, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1609008434, "tokens_per_second": 139375.47298112747, "data_wait_s": 8.352999429916963e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 11100, "loss": 0.7419378254562616, "lr": 0.0003, "grad_norm": 0.15071296947925078, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1623604035, "tokens_per_second": 148108.5150613146, "data_wait_s": 1.1514999641804025e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 11200, "loss": 0.7446037046611309, "lr": 0.0003, "grad_norm": 0.14735876864574618, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1638338285, "tokens_per_second": 148852.62966130654, "data_wait_s": 1.3658000170835294e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a85fadbd4a052813f58d2cef92a38590c46d3e26eb1f71b89afaf523a83f5981
-size 232885
+oid sha256:a4c8ad4db1b998516e3d98ec68748273326f1207ba6e6e75823f62851130fcc0
+size 234968