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": 77396.16768245464,
-      "required_pages_per_second_upper_estimate": 0.005354446464123062
+      "required_token_positions_per_second": 77389.9905237465,
+      "required_pages_per_second_upper_estimate": 0.005354019113950649
     },
     {
       "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": 9674.52096030683,
-      "required_pages_per_second_upper_estimate": 0.004050519899109566
+      "required_token_positions_per_second": 9673.748815468312,
+      "required_pages_per_second_upper_estimate": 0.004050196618189897
     },
     {
       "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": 19349.04192061366,
-      "required_pages_per_second_upper_estimate": 0.004029895163514306
+      "required_token_positions_per_second": 19347.497630936625,
+      "required_pages_per_second_upper_estimate": 0.004029573528700236
     },
     {
       "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": 29023.562880920494,
-      "required_pages_per_second_upper_estimate": 0.001739918278296668
+      "required_token_positions_per_second": 29021.24644640494,
+      "required_pages_per_second_upper_estimate": 0.0017397794115844513
     },
     {
       "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": 29023.562880920494,
-      "required_pages_per_second_upper_estimate": 0.0022564543528301026
+      "required_token_positions_per_second": 29021.24644640494,
+      "required_pages_per_second_upper_estimate": 0.002256274260235437
     },
     {
       "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": 29023.562880920494,
-      "required_pages_per_second_upper_estimate": 0.3722831015625825
+      "required_token_positions_per_second": 29021.24644640494,
+      "required_pages_per_second_upper_estimate": 0.37225338882781056
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.08816585614067662,
-    "producer_p95_seconds": 0.19003528600114805,
-    "gpu_update_seconds": 0.9526053060210343,
-    "producer_batches_per_second": 11.342259280105093,
-    "gpu_batches_per_second": 1.0497527083666267,
-    "producer_headroom": 10.804696372494428,
+    "producer_batch_seconds": 0.07182655081248868,
+    "producer_p95_seconds": 0.187659592998898,
+    "gpu_update_seconds": 0.9526813416184248,
+    "producer_batches_per_second": 13.92242824816429,
+    "gpu_batches_per_second": 1.0496689252895306,
+    "producer_headroom": 13.26363762204741,
     "sustainable": true,
-    "buffer_seconds": 60.01413427932516,
+    "buffer_seconds": 60.01892452196076,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 231394.41657960557,
-    "upload_bytes_per_second": 5887064.024184073,
+    "generation_bytes_per_second": 241406.69560121067,
+    "upload_bytes_per_second": 5824704.357963664,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 49.65185521165014,
+    "estimated_full_shards_per_hour": 51.80025721575966,
     "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": [
@@ -502,6 +502,12 @@
       "bytes": 10997251,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.0726386229998752,
+      "bytes": 2419938,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -69,3 +69,4 @@
 {"step": 491900, "loss": 0.6098073124885559, "lr": 0.0003, "grad_norm": 0.1502846487220614, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71997510019, "tokens_per_second": 144049.5154796702, "data_wait_s": 1.22540004667826e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 492000, "loss": 0.5890203714370728, "lr": 0.0003, "grad_norm": 0.1959452605804415, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72012187687, "tokens_per_second": 153246.40589248177, "data_wait_s": 9.343999408883974e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 492100, "loss": 0.579491138458252, "lr": 0.0003, "grad_norm": 0.16071308344928423, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72026911050, "tokens_per_second": 153743.89890390108, "data_wait_s": 9.590999979991466e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 492200, "loss": 0.5845812559127808, "lr": 0.0003, "grad_norm": 0.14824155212671003, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72041630681, "tokens_per_second": 153710.4138390692, "data_wait_s": 9.390998457092792e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fcdfc1f8a8111230d460669deafcd27e58bd2a9484d3ce0e9cf68ce146b2aea1
-size 151326
+oid sha256:84ab46682bf880feb62626cad9ebd8979a6de62b0cdafedd6f5a275b6a721ff1
+size 153444