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": 76134.7180230817,
-      "required_pages_per_second_upper_estimate": 0.0052671764497211745
+      "required_token_positions_per_second": 76189.62621202419,
+      "required_pages_per_second_upper_estimate": 0.005270975125636767
     },
     {
       "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": 9516.839752885213,
-      "required_pages_per_second_upper_estimate": 0.003984502070320178
+      "required_token_positions_per_second": 9523.703276503024,
+      "required_pages_per_second_upper_estimate": 0.003987375684332284
     },
     {
       "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": 19033.679505770426,
-      "required_pages_per_second_upper_estimate": 0.00396421348916861
+      "required_token_positions_per_second": 19047.406553006047,
+      "required_pages_per_second_upper_estimate": 0.00396707247110121
     },
     {
       "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": 28550.519258655637,
-      "required_pages_per_second_upper_estimate": 0.0017115600354376788
+      "required_token_positions_per_second": 28571.10982950907,
+      "required_pages_per_second_upper_estimate": 0.0017127944087203608
     },
     {
       "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": 28550.519258655637,
-      "required_pages_per_second_upper_estimate": 0.0022196772918980094
+      "required_token_positions_per_second": 28571.10982950907,
+      "required_pages_per_second_upper_estimate": 0.002221278118213514
     },
     {
       "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": 28550.519258655637,
-      "required_pages_per_second_upper_estimate": 0.3662154058908382
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28571.10982950907,
+      "required_pages_per_second_upper_estimate": 0.36647951962531355
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09799750696879528,
-    "producer_p95_seconds": 0.27451235000262386,
-    "gpu_update_seconds": 0.9683886919716173,
-    "producer_batches_per_second": 10.204341221847853,
-    "gpu_batches_per_second": 1.0326432023529961,
-    "producer_headroom": 9.881768648257298,
+    "producer_batch_seconds": 0.4531342907499436,
+    "producer_p95_seconds": 0.29220848299883073,
+    "gpu_update_seconds": 0.9676907955267577,
+    "producer_batches_per_second": 2.2068513030540813,
+    "gpu_batches_per_second": 1.0333879423288872,
+    "producer_headroom": 2.135549693061666,
     "sustainable": true,
-    "buffer_seconds": 60.04009890224027,
+    "buffer_seconds": 60.96452011818574,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 365489.8140163552,
+    "generation_bytes_per_second": 327027.87622578803,
     "upload_bytes_per_second": 5334191.644855798,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.42560592048638,
+    "estimated_full_shards_per_hour": 70.17256941871862,
     "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-v1/1791039222312944374-63c10d34/train.jsonl

@@ -307,3 +307,4 @@
 {"step": 635100, "loss": 0.5057151317596436, "lr": 0.0001, "grad_norm": 0.162308567881318, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92962080464, "tokens_per_second": 145506.0382751896, "data_wait_s": 1.3736000255448744e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 635200, "loss": 0.5750125050544739, "lr": 0.0001, "grad_norm": 0.1774653268777128, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92976760316, "tokens_per_second": 153024.1355079562, "data_wait_s": 1.801799953682348e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 635300, "loss": 0.5325800180435181, "lr": 0.0001, "grad_norm": 0.17676632920750399, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92991266546, "tokens_per_second": 150612.82307274116, "data_wait_s": 8.048002200666815e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 635400, "loss": 0.49283096194267273, "lr": 0.0001, "grad_norm": 0.18659582375256248, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93005764105, "tokens_per_second": 149815.13955222492, "data_wait_s": 1.4457997167482972e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:efbca1cd1196cc7457043938f02cea1aeec41cf59897f30811362541a1871eee
-size 660108
+oid sha256:50036c5b1f0ac3bd8c5a26103d68a5725d0d914b2f6481a0aeea9ed2f8b8be4d
+size 662226