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": 75733.47975957241,
-      "required_pages_per_second_upper_estimate": 0.00656438608081013
+      "required_token_positions_per_second": 75821.39814967013,
+      "required_pages_per_second_upper_estimate": 0.006572006624036678
     },
     {
       "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": 9466.684969946551,
-      "required_pages_per_second_upper_estimate": 0.003776402174065163
+      "required_token_positions_per_second": 9477.674768708766,
+      "required_pages_per_second_upper_estimate": 0.003780786169103545
     },
     {
       "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": 18933.369939893102,
-      "required_pages_per_second_upper_estimate": 0.004379901919664064
+      "required_token_positions_per_second": 18955.349537417533,
+      "required_pages_per_second_upper_estimate": 0.0043849865127236366
     },
     {
       "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": 28400.05490983965,
-      "required_pages_per_second_upper_estimate": 0.0017294143404367258
+      "required_token_positions_per_second": 28433.0243061263,
+      "required_pages_per_second_upper_estimate": 0.0017314220036935286
     },
     {
       "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": 28400.05490983965,
-      "required_pages_per_second_upper_estimate": 0.0021863148433431193
+      "required_token_positions_per_second": 28433.0243061263,
+      "required_pages_per_second_upper_estimate": 0.0021888529187343957
     },
     {
       "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": 28400.05490983965,
-      "required_pages_per_second_upper_estimate": 0.3526336331107398
+      "required_token_positions_per_second": 28433.0243061263,
+      "required_pages_per_second_upper_estimate": 0.35304300267114863
     }
   ],
   "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": 0.20705526900795235,
-    "producer_p95_seconds": 0.26620429800095735,
-    "gpu_update_seconds": 0.973519244514591,
-    "producer_batches_per_second": 4.829628363437557,
-    "gpu_batches_per_second": 1.0272010601070476,
-    "producer_headroom": 4.7017361556599715,
+    "producer_batch_seconds": 0.07976260507811617,
+    "producer_p95_seconds": 0.24803905700173345,
+    "gpu_update_seconds": 0.972390404282208,
+    "producer_batches_per_second": 12.537203355139187,
+    "gpu_batches_per_second": 1.0283935295907949,
+    "producer_headroom": 12.191056239072049,
     "sustainable": true,
-    "buffer_seconds": 60.358193159904644,
+    "buffer_seconds": 60.288205065496896,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 492270.96945209627,
-    "upload_bytes_per_second": 5845659.336047131,
+    "generation_bytes_per_second": 473590.56530836294,
+    "upload_bytes_per_second": 5948249.745275737,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.62989056274573,
+    "estimated_full_shards_per_hour": 101.62151069105307,
     "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": [
@@ -406,6 +406,12 @@
       "bytes": 5851026,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.3481963590020314,
+      "bytes": 12258822,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -93,3 +93,4 @@
 {"step": 94300, "loss": 0.6212357766926289, "lr": 0.0003, "grad_norm": 0.13607407178389822, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13800147761, "tokens_per_second": 147773.20663485298, "data_wait_s": 9.108000085689127e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 94400, "loss": 0.6433022990822792, "lr": 0.0003, "grad_norm": 0.13999482436726243, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13814687815, "tokens_per_second": 148644.52013275062, "data_wait_s": 9.969000529963523e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 94500, "loss": 0.6318197883665562, "lr": 0.0003, "grad_norm": 0.15126675226290945, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13829351755, "tokens_per_second": 149767.10950509724, "data_wait_s": 1.0874999134102836e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 94600, "loss": 0.6887232102453709, "lr": 0.0003, "grad_norm": 0.14319270546615992, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13844060383, "tokens_per_second": 149659.1972642344, "data_wait_s": 1.006700040306896e-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 sha256:70cda0b97281d0004bf27bc0f04726bd2c51f7889f6549970c3d6621551d78a2
-size 202158
+oid sha256:46e00877c9130485fb3883ceaadf00cb770a70bf90df745304f63d3f081c70e6
+size 204276