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": 75061.23950953798,
-      "required_pages_per_second_upper_estimate": 0.006506117999714483
+      "required_token_positions_per_second": 75410.11962323569,
+      "required_pages_per_second_upper_estimate": 0.006536358043741236
     },
     {
       "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": 9382.654938692247,
-      "required_pages_per_second_upper_estimate": 0.0037428813382368946
+      "required_token_positions_per_second": 9426.264952904461,
+      "required_pages_per_second_upper_estimate": 0.003760278024933964
     },
     {
       "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": 18765.309877384494,
-      "required_pages_per_second_upper_estimate": 0.004341024181958778
+      "required_token_positions_per_second": 18852.529905808922,
+      "required_pages_per_second_upper_estimate": 0.004361201000514696
     },
     {
       "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": 28147.96481607674,
-      "required_pages_per_second_upper_estimate": 0.0017140633763410692
+      "required_token_positions_per_second": 28278.794858713383,
+      "required_pages_per_second_upper_estimate": 0.001722030239525456
     },
     {
       "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": 28147.96481607674,
-      "required_pages_per_second_upper_estimate": 0.0021669082500952075
+      "required_token_positions_per_second": 28278.794858713383,
+      "required_pages_per_second_upper_estimate": 0.002176979908938114
     },
     {
       "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": 28147.96481607674,
-      "required_pages_per_second_upper_estimate": 0.3495035178374752
+      "required_token_positions_per_second": 28278.794858713383,
+      "required_pages_per_second_upper_estimate": 0.3511279891070363
     }
   ],
   "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": 0.3279157915469071,
-    "producer_p95_seconds": 0.2065597950004303,
-    "gpu_update_seconds": 0.9822379763743635,
-    "producer_batches_per_second": 3.049563411638728,
-    "gpu_batches_per_second": 1.018083218174072,
-    "producer_headroom": 2.9953969942733245,
+    "producer_batch_seconds": 0.04961033224998346,
+    "producer_p95_seconds": 0.17037309699935577,
+    "gpu_update_seconds": 0.9776937149597441,
+    "producer_batches_per_second": 20.15709136881125,
+    "gpu_batches_per_second": 1.022815207563418,
+    "producer_headroom": 19.707461543156064,
     "sustainable": true,
-    "buffer_seconds": 60.898754535210536,
+    "buffer_seconds": 60.61701032750413,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 399669.8161313674,
+    "generation_bytes_per_second": 399136.7533068909,
     "upload_bytes_per_second": 4853717.286250679,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.75983870464103,
+    "estimated_full_shards_per_hour": 85.64545583157583,
     "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-v1/1790563279934888633-82973089/train.jsonl

@@ -42,3 +42,4 @@
 {"step": 159200, "loss": 0.5842249542474747, "lr": 0.0003, "grad_norm": 0.14998375597947464, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23299799851, "tokens_per_second": 147627.9249082167, "data_wait_s": 1.2814000001526438e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 159300, "loss": 0.5889497511088848, "lr": 0.0003, "grad_norm": 0.14231023699674406, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23314483323, "tokens_per_second": 149607.7924737093, "data_wait_s": 1.3675999980478082e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 159400, "loss": 0.5490333624184132, "lr": 0.0003, "grad_norm": 0.14879176756412074, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23329089121, "tokens_per_second": 148774.9649715777, "data_wait_s": 1.605200031917775e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 159500, "loss": 0.5589956678450108, "lr": 0.0003, "grad_norm": 0.14677640845151912, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23343720107, "tokens_per_second": 149320.57667826503, "data_wait_s": 1.1400999937904999e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fb318ee0ce100beea4b441d3d604b75c4ce8b8b299104a2ae88cfc64abdc402b
-size 93357
+oid sha256:62db26e72ffc3de219045017df18b84d24bc60cb17c80d1e27b9109fb3ddbdc5
+size 95475