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": 75681.78761690138,
-      "required_pages_per_second_upper_estimate": 0.006559905536895943
+      "required_token_positions_per_second": 75757.47971483711,
+      "required_pages_per_second_upper_estimate": 0.0065664663361051446
     },
     {
       "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": 9460.223452112672,
-      "required_pages_per_second_upper_estimate": 0.003773824577992928
+      "required_token_positions_per_second": 9469.68496435464,
+      "required_pages_per_second_upper_estimate": 0.003777598916688463
     },
     {
       "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": 18920.446904225344,
-      "required_pages_per_second_upper_estimate": 0.004376912402799988
+      "required_token_positions_per_second": 18939.36992870928,
+      "required_pages_per_second_upper_estimate": 0.00438128991148049
     },
     {
       "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": 28380.670356338018,
-      "required_pages_per_second_upper_estimate": 0.0017282339228313684
+      "required_token_positions_per_second": 28409.05489306392,
+      "required_pages_per_second_upper_estimate": 0.001729962392195821
     },
     {
       "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": 28380.670356338018,
-      "required_pages_per_second_upper_estimate": 0.002184822566754685
+      "required_token_positions_per_second": 28409.05489306392,
+      "required_pages_per_second_upper_estimate": 0.0021870076869652715
     },
     {
       "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": 28380.670356338018,
-      "required_pages_per_second_upper_estimate": 0.35239294183217673
+      "required_token_positions_per_second": 28409.05489306392,
+      "required_pages_per_second_upper_estimate": 0.352745382781379
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.1251538520076565,
-    "producer_p95_seconds": 0.3769494529988151,
-    "gpu_update_seconds": 0.9741841772185484,
-    "producer_batches_per_second": 7.990165575877148,
-    "gpu_batches_per_second": 1.0264999405504203,
-    "producer_headroom": 7.783892877375848,
+    "producer_batch_seconds": 0.10865640378898433,
+    "producer_p95_seconds": 0.3665677380049601,
+    "gpu_update_seconds": 0.9732108338018055,
+    "producer_batches_per_second": 9.20332318325246,
+    "gpu_batches_per_second": 1.0275265803336198,
+    "producer_headroom": 8.956773828920614,
     "sustainable": true,
-    "buffer_seconds": 60.39941898755001,
+    "buffer_seconds": 60.339071695711944,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 429635.7499050022,
-    "upload_bytes_per_second": 5936969.930406086,
+    "generation_bytes_per_second": 452758.7132928658,
+    "upload_bytes_per_second": 5865395.460259779,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.1898305212264,
+    "estimated_full_shards_per_hour": 97.1514801892231,
     "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": 1.0866164300023229,
-      "bytes": 7550162,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.2529126609988452,
       "bytes": 5669267,
@@ -562,6 +556,12 @@
       "bytes": 7665641,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 0.8218483260061475,
+      "bytes": 877627,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -311,3 +311,4 @@
 {"step": 116100, "loss": 0.5674298200756311, "lr": 0.0003, "grad_norm": 0.14025242568622506, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16989562756, "tokens_per_second": 148254.01830830792, "data_wait_s": 8.544993761461228e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 116200, "loss": 0.63331014290452, "lr": 0.0003, "grad_norm": 0.14362439571118016, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17004101407, "tokens_per_second": 148138.27393720526, "data_wait_s": 1.3131000741850585e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 116300, "loss": 0.6128046456724405, "lr": 0.0003, "grad_norm": 0.15431430011580516, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17018720502, "tokens_per_second": 148930.9296571331, "data_wait_s": 2.039800165221095e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 116400, "loss": 0.5653275307267904, "lr": 0.0003, "grad_norm": 0.13909962304253426, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17033279015, "tokens_per_second": 133019.79481446205, "data_wait_s": 1.032499858411029e-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 sha256:7a78dcaf3029a4bfd4f624554d8818cdf5d2a541995f9930d23d9ac6683d265f
-size 667797
+oid sha256:39810110ae9f5e7edda322f81c0da99a0919ab28c4669b1b8885aada6f6c4dca
+size 669915