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": 74553.95244752878,
-      "required_pages_per_second_upper_estimate": 0.006462147642886829
+      "required_token_positions_per_second": 74047.19997882469,
+      "required_pages_per_second_upper_estimate": 0.006418223623251949
     },
     {
       "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": 9319.244055941097,
-      "required_pages_per_second_upper_estimate": 0.0037175857890302763
+      "required_token_positions_per_second": 9255.899997353086,
+      "required_pages_per_second_upper_estimate": 0.003692316896981445
     },
     {
       "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": 18638.488111882194,
-      "required_pages_per_second_upper_estimate": 0.004311686198496662
+      "required_token_positions_per_second": 18511.799994706173,
+      "required_pages_per_second_upper_estimate": 0.0042823791322227
     },
     {
       "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": 27957.73216782329,
-      "required_pages_per_second_upper_estimate": 0.0017024792061360045
+      "required_token_positions_per_second": 27767.699992059257,
+      "required_pages_per_second_upper_estimate": 0.001690907243653747
     },
     {
       "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": 27957.73216782329,
-      "required_pages_per_second_upper_estimate": 0.002152263614234983
+      "required_token_positions_per_second": 27767.699992059257,
+      "required_pages_per_second_upper_estimate": 0.0021376344113019368
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27957.73216782329,
-      "required_pages_per_second_upper_estimate": 0.34714146501388543
+      "required_token_positions_per_second": 27767.699992059257,
+      "required_pages_per_second_upper_estimate": 0.34478190138767595
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08155526450769912,
-    "producer_p95_seconds": 0.22470485200028634,
-    "gpu_update_seconds": 0.9889214130114686,
-    "producer_batches_per_second": 12.261624139611444,
-    "gpu_batches_per_second": 1.0112026970422197,
-    "producer_headroom": 12.125782669960083,
+    "producer_batch_seconds": 0.10681534571889983,
+    "producer_p95_seconds": 0.2374221589998342,
+    "gpu_update_seconds": 0.9956892363395784,
+    "producer_batches_per_second": 9.36195069416005,
+    "gpu_batches_per_second": 1.0043294267961247,
+    "producer_headroom": 9.321593537317005,
     "sustainable": true,
-    "buffer_seconds": 60.32420619369959,
+    "buffer_seconds": 60.73704341671428,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 424673.0058353111,
-    "upload_bytes_per_second": 5998250.768080359,
+    "generation_bytes_per_second": 471450.6373317519,
+    "upload_bytes_per_second": 5991865.635180688,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.12494117063999,
+    "estimated_full_shards_per_hour": 101.1623319622461,
     "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": 0.9057083049992798,
-      "bytes": 1081901,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.247454252999887,
       "bytes": 13520201,
@@ -562,6 +556,12 @@
       "bytes": 5991225,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 0.9874920929978543,
+      "bytes": 1116876,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -134,3 +134,4 @@
 {"step": 98400, "loss": 0.6232114732265472, "lr": 0.0003, "grad_norm": 0.1266015312917367, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14398943189, "tokens_per_second": 147535.5966096913, "data_wait_s": 1.2119999155402184e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 98500, "loss": 0.5763075035065413, "lr": 0.0003, "grad_norm": 0.1383303369129346, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14413621505, "tokens_per_second": 149235.90115195137, "data_wait_s": 1.461600186303258e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 98600, "loss": 0.6116198264062405, "lr": 0.0003, "grad_norm": 0.13578267294458274, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14428372687, "tokens_per_second": 149459.39880108126, "data_wait_s": 1.2582997442223132e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 98700, "loss": 0.6326380763202906, "lr": 0.0003, "grad_norm": 0.13642060506661843, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14442954752, "tokens_per_second": 148356.8771632658, "data_wait_s": 1.5718000213382766e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:be104c3575b1cb3f1d6bb1199b028dd2eea3f46722c9d35f5200ff62344304ed
-size 289779
+oid sha256:41a6598948a34dcd2f4d1612f3b41a82110237d347d4f70c2eedd1d9841411a9
+size 291897