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": 76998.98314704961,
-      "required_pages_per_second_upper_estimate": 0.005326968316368649
+      "required_token_positions_per_second": 77088.38450118755,
+      "required_pages_per_second_upper_estimate": 0.005333153309487633
     },
     {
       "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": 9624.872893381202,
-      "required_pages_per_second_upper_estimate": 0.004029733290257338
+      "required_token_positions_per_second": 9636.048062648444,
+      "required_pages_per_second_upper_estimate": 0.004034412100265461
     },
     {
       "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": 19249.745786762403,
-      "required_pages_per_second_upper_estimate": 0.004009214397448232
+      "required_token_positions_per_second": 19272.09612529689,
+      "required_pages_per_second_upper_estimate": 0.0040138693835469015
     },
     {
       "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": 28874.618680143605,
-      "required_pages_per_second_upper_estimate": 0.00173098930088472
+      "required_token_positions_per_second": 28908.144187945334,
+      "required_pages_per_second_upper_estimate": 0.001732999103886947
     },
     {
       "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": 28874.618680143605,
-      "required_pages_per_second_upper_estimate": 0.002244874596356002
+      "required_token_positions_per_second": 28908.144187945334,
+      "required_pages_per_second_upper_estimate": 0.002247481057124462
     },
     {
       "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": 28874.618680143605,
-      "required_pages_per_second_upper_estimate": 0.3703726052788395
+      "required_token_positions_per_second": 28908.144187945334,
+      "required_pages_per_second_upper_estimate": 0.37080263449603434
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07270612068749838,
-    "producer_p95_seconds": 0.12818421899999066,
-    "gpu_update_seconds": 0.9575191383917003,
-    "producer_batches_per_second": 13.754000220946287,
-    "gpu_batches_per_second": 1.0443655483269534,
-    "producer_headroom": 13.169718440999743,
+    "producer_batch_seconds": 0.06244436849218804,
+    "producer_p95_seconds": 0.12829015700003765,
+    "gpu_update_seconds": 0.9564086791683151,
+    "producer_batches_per_second": 16.01425435385903,
+    "gpu_batches_per_second": 1.0455781317977912,
+    "producer_headroom": 15.316171854439755,
     "sustainable": true,
-    "buffer_seconds": 60.32370571867712,
+    "buffer_seconds": 60.25374678760385,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 59,
+    "upload_pages_per_commit": 60,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 337361.5618755913,
-    "upload_bytes_per_second": 2663270.868004823,
+    "generation_bytes_per_second": 386265.1001743187,
+    "upload_bytes_per_second": 2715257.5497594303,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.3899378032761,
+    "estimated_full_shards_per_hour": 82.88349870607539,
     "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": [
@@ -220,6 +220,12 @@
       "bytes": 2710887,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 2.055465744000003,
+      "bytes": 6120157,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -5,3 +5,4 @@
 {"step": 843000, "loss": 0.5480272769927979, "lr": 3e-05, "grad_norm": 0.30482483918165887, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123391515182, "tokens_per_second": 139028.7191700509, "data_wait_s": 9.363000003759225e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 843100, "loss": 0.47235390543937683, "lr": 3e-05, "grad_norm": 0.30681393268363527, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123406135026, "tokens_per_second": 151282.84953002178, "data_wait_s": 1.0688999964258983e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 843200, "loss": 0.5120399594306946, "lr": 3e-05, "grad_norm": 0.29002387894380416, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123420670414, "tokens_per_second": 150754.8370364941, "data_wait_s": 1.2794999975085375e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 843300, "loss": 0.5344165563583374, "lr": 3e-05, "grad_norm": 0.298620691567682, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123435375990, "tokens_per_second": 151886.9994461236, "data_wait_s": 6.656999858023482e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bb14784531c32d13a9109a241018dcd53985349d8c7884d44c486054b27d3eba
-size 14991
+oid sha256:22dbdfd32f699d7b9c9b2650bf6e0692d00ce8f92ea70589982b1376dbd396bf
+size 17109