Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76695.66113094748,
-      "required_pages_per_second_upper_estimate": 0.0053059837954906605
+      "required_token_positions_per_second": 77190.12359386288,
+      "required_pages_per_second_upper_estimate": 0.00534019185598613
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9586.957641368435,
-      "required_pages_per_second_upper_estimate": 0.0040138589660000884
+      "required_token_positions_per_second": 9648.76544923286,
+      "required_pages_per_second_upper_estimate": 0.0040397366044591255
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19173.91528273687,
-      "required_pages_per_second_upper_estimate": 0.003993420903244584
+      "required_token_positions_per_second": 19297.53089846572,
+      "required_pages_per_second_upper_estimate": 0.004019166776037894
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28760.872924105304,
-      "required_pages_per_second_upper_estimate": 0.0017241704164899369
+      "required_token_positions_per_second": 28946.29634769858,
+      "required_pages_per_second_upper_estimate": 0.0017352862676091793
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28760.872924105304,
-      "required_pages_per_second_upper_estimate": 0.00223603136413873
+      "required_token_positions_per_second": 28946.29634769858,
+      "required_pages_per_second_upper_estimate": 0.002250447219731661
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28760.872924105304,
-      "required_pages_per_second_upper_estimate": 0.3689135968510576
+      "required_token_positions_per_second": 28946.29634769858,
+      "required_pages_per_second_upper_estimate": 0.37129200943675145
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11237955500013186,
-    "producer_p95_seconds": 0.347817492998729,
-    "gpu_update_seconds": 0.961306010181194,
-    "producer_batches_per_second": 8.898415730502107,
-    "gpu_batches_per_second": 1.0402514801832068,
-    "producer_headroom": 8.554100522822555,
+    "producer_batch_seconds": 0.36914079823446855,
+    "producer_p95_seconds": 0.35453830500046024,
+    "gpu_update_seconds": 0.9551481014322648,
+    "producer_batches_per_second": 2.7089934376877687,
+    "gpu_batches_per_second": 1.046958056557385,
+    "producer_headroom": 2.5874899387999366,
     "sustainable": true,
-    "buffer_seconds": 60.56227864141522,
+    "buffer_seconds": 60.17433039023268,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 539811.4523555023,
-    "upload_bytes_per_second": 5296327.716165321,
+    "generation_bytes_per_second": 476081.46501432825,
+    "upload_bytes_per_second": 5404763.012242273,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 115.8309715080147,
+    "estimated_full_shards_per_hour": 102.15599978277574,
     "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": 1.1440478279982926,
-      "bytes": 2552467,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.1846777530008694,
       "bytes": 10218604,
@@ -562,6 +556,12 @@
       "bytes": 5828706,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.4654938949970528,
+      "bytes": 13087295,
+      "pages": 32,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -366,3 +366,4 @@
 {"step": 433700, "loss": 0.5695508122444153, "lr": 0.0003, "grad_norm": 0.15051090979827284, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63480987854, "tokens_per_second": 151681.67218986698, "data_wait_s": 1.4263998309616e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 433800, "loss": 0.5868895649909973, "lr": 0.0003, "grad_norm": 0.15602795269241654, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63495695720, "tokens_per_second": 153434.10422555634, "data_wait_s": 1.603399869054556e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 433900, "loss": 0.637799084186554, "lr": 0.0003, "grad_norm": 0.1673257747400092, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63510307268, "tokens_per_second": 151849.88024531046, "data_wait_s": 1.6525998944416642e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 434000, "loss": 0.591548502445221, "lr": 0.0003, "grad_norm": 0.1615272696927658, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63524954041, "tokens_per_second": 151112.94545203622, "data_wait_s": 1.4924997230991721e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0b2c6ab308c9f283da7b6ca60d7e0ea71201874518933e88ecf042daf0f1b0b2
-size 785070
+oid sha256:438214fde4a0659e3e74ae4773a004c344a2767e9685e0e88a5e952b4fd8c97b
+size 787188