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": 77691.91904373578,
-      "required_pages_per_second_upper_estimate": 0.005374907229534194
+      "required_token_positions_per_second": 77145.51281026368,
+      "required_pages_per_second_upper_estimate": 0.005337105578465459
     },
     {
       "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": 9711.489880466972,
-      "required_pages_per_second_upper_estimate": 0.004065998013981777
+      "required_token_positions_per_second": 9643.18910128296,
+      "required_pages_per_second_upper_estimate": 0.004037401904019889
     },
     {
       "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": 19422.979760933944,
-      "required_pages_per_second_upper_estimate": 0.004045294465781048
+      "required_token_positions_per_second": 19286.37820256592,
+      "required_pages_per_second_upper_estimate": 0.004016843963598335
     },
     {
       "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": 29134.469641400916,
-      "required_pages_per_second_upper_estimate": 0.0017465669692426516
+      "required_token_positions_per_second": 28929.567303848882,
+      "required_pages_per_second_upper_estimate": 0.001734283386974161
     },
     {
       "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": 29134.469641400916,
-      "required_pages_per_second_upper_estimate": 0.0022650768656301713
+      "required_token_positions_per_second": 28929.567303848882,
+      "required_pages_per_second_upper_estimate": 0.0022491466101556344
     },
     {
       "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": 29134.469641400916,
-      "required_pages_per_second_upper_estimate": 0.37370569440362383
+      "required_token_positions_per_second": 28929.567303848882,
+      "required_pages_per_second_upper_estimate": 0.3710774272244953
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07168984638289544,
-    "producer_p95_seconds": 0.2536663499995484,
-    "gpu_update_seconds": 0.9489790046053009,
-    "producer_batches_per_second": 13.948976744335598,
-    "gpu_batches_per_second": 1.0537640929326142,
-    "producer_headroom": 13.237286066102088,
+    "producer_batch_seconds": 0.3951263455080607,
+    "producer_p95_seconds": 0.2589216199994553,
+    "gpu_update_seconds": 0.9557004330417905,
+    "producer_batches_per_second": 2.5308360512235186,
+    "gpu_batches_per_second": 1.0463529840801824,
+    "producer_headroom": 2.418721110112092,
     "sustainable": true,
-    "buffer_seconds": 60.73465629473926,
+    "buffer_seconds": 60.2091272816328,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 317370.00599035923,
-    "upload_bytes_per_second": 5511058.666317652,
+    "generation_bytes_per_second": 259219.26221638508,
+    "upload_bytes_per_second": 5422540.174717165,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.10021528990825,
+    "estimated_full_shards_per_hour": 55.62241935604729,
     "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": 1.495498950000183,
-      "bytes": 7289548,
-      "pages": 26,
-      "shards": 3
-    },
     {
       "seconds": 1.2270069759997568,
       "bytes": 11800938,
@@ -562,6 +556,12 @@
       "bytes": 6001693,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.5835260300009395,
+      "bytes": 761598,
+      "pages": 23,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -239,3 +239,4 @@
 {"step": 708900, "loss": 0.5142809748649597, "lr": 8e-05, "grad_norm": 0.19966561877811448, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103762859270, "tokens_per_second": 150353.90736413703, "data_wait_s": 1.4174001989886165e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 709000, "loss": 0.5492202639579773, "lr": 8e-05, "grad_norm": 0.2270388262767895, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103777427626, "tokens_per_second": 152381.12924728592, "data_wait_s": 1.1015999916708097e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 709100, "loss": 0.5522074103355408, "lr": 8e-05, "grad_norm": 0.21061085954004322, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103792103834, "tokens_per_second": 118702.04479120397, "data_wait_s": 1.6681999113643542e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 709200, "loss": 0.5702271461486816, "lr": 8e-05, "grad_norm": 0.23457596407394352, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103806820972, "tokens_per_second": 152595.79665767297, "data_wait_s": 1.3566997949965298e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:af41386e3ecce6e0a99b56e26ee2437e7ea435e5fc078d561b52d9c13204b297
-size 513735
+oid sha256:bb78b81f3b09e220c15f355df0e3d938aebe44f0ac3edaab790c9a105903e8be
+size 515853