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": 75756.23144108093,
-      "required_pages_per_second_upper_estimate": 0.005240991869768497
+      "required_token_positions_per_second": 77155.50268032493,
+      "required_pages_per_second_upper_estimate": 0.005337796700855988
     },
     {
       "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": 9469.528930135117,
-      "required_pages_per_second_upper_estimate": 0.003964694016797036
+      "required_token_positions_per_second": 9644.437835040617,
+      "required_pages_per_second_upper_estimate": 0.004037924722767694
     },
     {
       "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": 18939.057860270234,
-      "required_pages_per_second_upper_estimate": 0.0039445062957515166
+      "required_token_positions_per_second": 19288.875670081234,
+      "required_pages_per_second_upper_estimate": 0.004017364120219128
     },
     {
       "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": 28408.586790405352,
-      "required_pages_per_second_upper_estimate": 0.0017030514006844048
+      "required_token_positions_per_second": 28933.31350512185,
+      "required_pages_per_second_upper_estimate": 0.0017345079660203572
     },
     {
       "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": 28408.586790405352,
-      "required_pages_per_second_upper_estimate": 0.0022086426667865
+      "required_token_positions_per_second": 28933.31350512185,
+      "required_pages_per_second_upper_estimate": 0.0022494378608268095
     },
     {
       "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": 28408.586790405352,
-      "required_pages_per_second_upper_estimate": 0.36439484858333465
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28933.31350512185,
+      "required_pages_per_second_upper_estimate": 0.3711254794720674
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.11900041863310662,
+    "producer_batch_seconds": 0.12189683281286534,
     "producer_p95_seconds": 0.3881533629974001,
-    "gpu_update_seconds": 0.973226869889134,
-    "producer_batches_per_second": 8.403331782244623,
-    "gpu_batches_per_second": 1.0275096495372307,
-    "producer_headroom": 8.178348287073812,
+    "gpu_update_seconds": 0.9555766917296106,
+    "producer_batches_per_second": 8.203658593289203,
+    "gpu_batches_per_second": 1.0464884803646501,
+    "producer_headroom": 7.839224938654487,
     "sustainable": true,
-    "buffer_seconds": 60.340065933126304,
+    "buffer_seconds": 60.201331578965465,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 286542.3759552319,
-    "upload_bytes_per_second": 4425533.958861365,
+    "generation_bytes_per_second": 274206.6202150226,
+    "upload_bytes_per_second": 4471118.07163161,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.4853235148689,
+    "estimated_full_shards_per_hour": 58.838357494716725,
     "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": 1.328966004999529,
-      "bytes": 1184587,
-      "pages": 36,
-      "shards": 1
-    },
     {
       "seconds": 2.0095185120007955,
       "bytes": 12961722,
@@ -562,6 +556,12 @@
       "bytes": 13040260,
       "pages": 38,
       "shards": 4
+    },
+    {
+      "seconds": 1.5299789529963164,
+      "bytes": 6406874,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -422,3 +422,4 @@
 {"step": 842200, "loss": 0.47381365299224854, "lr": 3e-05, "grad_norm": 0.29832004624987274, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123274331554, "tokens_per_second": 151585.23609351562, "data_wait_s": 1.1219999578315765e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 842300, "loss": 0.5562163591384888, "lr": 3e-05, "grad_norm": 0.2983377293134837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123289015698, "tokens_per_second": 151895.5803168626, "data_wait_s": 7.576003554277122e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 842400, "loss": 0.5182906985282898, "lr": 3e-05, "grad_norm": 0.2880123845292042, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123303506551, "tokens_per_second": 150064.32580190475, "data_wait_s": 1.2801996490452439e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 842500, "loss": 0.5224790573120117, "lr": 3e-05, "grad_norm": 0.29719003755818924, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123318031370, "tokens_per_second": 126679.53084258523, "data_wait_s": 1.3865996152162552e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2159469e9d0b0485e6b3e009cc5c979cc1179b511cf1f53104ef77923c179660
-size 904461
+oid sha256:ff98bef82689225b2c09aa389e60a8efa235c1406d259c03fb408fad3d900ae1
+size 906579