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": 75679.23702373239,
-      "required_pages_per_second_upper_estimate": 0.0065596844579444745
+      "required_token_positions_per_second": 75288.42863405673,
+      "required_pages_per_second_upper_estimate": 0.006525810177222196
     },
     {
       "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": 9459.904627966549,
-      "required_pages_per_second_upper_estimate": 0.003773697394274194
+      "required_token_positions_per_second": 9411.053579257092,
+      "required_pages_per_second_upper_estimate": 0.0037542099805557253
     },
     {
       "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": 18919.809255933098,
-      "required_pages_per_second_upper_estimate": 0.004376764894089804
+      "required_token_positions_per_second": 18822.107158514184,
+      "required_pages_per_second_upper_estimate": 0.004354163233350124
     },
     {
       "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": 28379.71388389965,
-      "required_pages_per_second_upper_estimate": 0.0017281756786780948
+      "required_token_positions_per_second": 28233.16073777128,
+      "required_pages_per_second_upper_estimate": 0.0017192513609838108
     },
     {
       "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": 28379.71388389965,
-      "required_pages_per_second_upper_estimate": 0.002184748934858694
+      "required_token_positions_per_second": 28233.16073777128,
+      "required_pages_per_second_upper_estimate": 0.0021734668679846587
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28379.71388389965,
-      "required_pages_per_second_upper_estimate": 0.35238106564559957
+      "required_token_positions_per_second": 28233.16073777128,
+      "required_pages_per_second_upper_estimate": 0.3505613660525135
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.1031526438279684,
-    "producer_p95_seconds": 0.3958275359982508,
-    "gpu_update_seconds": 0.9742170098369186,
-    "producer_batches_per_second": 9.6943710106717,
-    "gpu_batches_per_second": 1.0264653459165094,
-    "producer_headroom": 9.444421138266291,
+    "producer_batch_seconds": 0.5678702997424239,
+    "producer_p95_seconds": 0.4001976889994694,
+    "gpu_update_seconds": 0.9792739912046606,
+    "producer_batches_per_second": 1.7609654888688187,
+    "gpu_batches_per_second": 1.0211646678881394,
+    "producer_headroom": 1.7244677026582345,
     "sustainable": true,
-    "buffer_seconds": 60.40145460988895,
+    "buffer_seconds": 60.71498745468896,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 243638.5247957689,
+    "generation_bytes_per_second": 222546.10126369406,
     "upload_bytes_per_second": 5760356.207952416,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 52.27915580658722,
+    "estimated_full_shards_per_hour": 47.753212723094144,
     "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-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -382,3 +382,4 @@
 {"step": 153200, "loss": 0.5679905088618398, "lr": 0.0003, "grad_norm": 0.15123643265815698, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22420385300, "tokens_per_second": 119521.99032782183, "data_wait_s": 1.469699782319367e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 153300, "loss": 0.6331153865903616, "lr": 0.0003, "grad_norm": 0.18648370077675916, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22435097611, "tokens_per_second": 150503.97369121175, "data_wait_s": 1.3190998288337141e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 153400, "loss": 0.6494401935487986, "lr": 0.0003, "grad_norm": 0.15229083702367008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22449860820, "tokens_per_second": 140538.03353914976, "data_wait_s": 1.2790995242539793e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 153500, "loss": 0.6256589218974113, "lr": 0.0003, "grad_norm": 0.1370351269445419, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22464484527, "tokens_per_second": 148875.36384478706, "data_wait_s": 1.335500564891845e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d2c98a110fb497eae23efa13d757b69639d8d801b84ddd7603e51c6f2f627914
-size 818958
+oid sha256:ce7e2d0358d9e3d49d248dd649e7cd0846153194edb04cf7c4d11e6d31248bf6
+size 821076