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": 76001.69535208859,
-      "required_pages_per_second_upper_estimate": 0.006587634328583228
+      "required_token_positions_per_second": 76037.91405531652,
+      "required_pages_per_second_upper_estimate": 0.006590773673983561
     },
     {
       "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": 9500.211919011073,
-      "required_pages_per_second_upper_estimate": 0.00378977657531956
+      "required_token_positions_per_second": 9504.739256914565,
+      "required_pages_per_second_upper_estimate": 0.003791582598098997
     },
     {
       "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": 19000.423838022147,
-      "required_pages_per_second_upper_estimate": 0.004395413658887917
+      "required_token_positions_per_second": 19009.47851382913,
+      "required_pages_per_second_upper_estimate": 0.004397508298779012
     },
     {
       "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": 28500.635757033222,
-      "required_pages_per_second_upper_estimate": 0.0017355391863238922
+      "required_token_positions_per_second": 28514.217770743697,
+      "required_pages_per_second_upper_estimate": 0.0017363662596995414
     },
     {
       "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": 28500.635757033222,
-      "required_pages_per_second_upper_estimate": 0.0021940578353856852
+      "required_token_positions_per_second": 28514.217770743697,
+      "required_pages_per_second_upper_estimate": 0.002195103416398538
     },
     {
       "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": 28500.635757033222,
-      "required_pages_per_second_upper_estimate": 0.3538825106104427
+      "required_token_positions_per_second": 28514.217770743697,
+      "required_pages_per_second_upper_estimate": 0.354051153764651
     }
   ],
   "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": 0.0951726115389846,
-    "producer_p95_seconds": 0.26733814100043674,
-    "gpu_update_seconds": 0.970083623246095,
-    "producer_batches_per_second": 10.507224545271411,
-    "gpu_batches_per_second": 1.0308389669065836,
-    "producer_headroom": 10.192886457137192,
+    "producer_batch_seconds": 0.14274976447651966,
+    "producer_p95_seconds": 0.24199199400027283,
+    "gpu_update_seconds": 0.9696215488810478,
+    "producer_batches_per_second": 7.005265498455418,
+    "gpu_batches_per_second": 1.031330214508959,
+    "producer_headroom": 6.792456382935308,
     "sustainable": true,
-    "buffer_seconds": 60.14518464125789,
+    "buffer_seconds": 60.116536030624964,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 404369.37813788373,
-    "upload_bytes_per_second": 5161757.108373739,
+    "generation_bytes_per_second": 409243.6114292342,
+    "upload_bytes_per_second": 5406284.695430953,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.768255311035,
+    "estimated_full_shards_per_hour": 87.81415230901499,
     "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": 3.5350073540002995,
-      "bytes": 5274454,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 1.0228857370002515,
       "bytes": 6079721,
@@ -562,6 +556,12 @@
       "bytes": 2309383,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.056229023000924,
+      "bytes": 11737562,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -114,3 +114,4 @@
 {"step": 166400, "loss": 0.6067061927169561, "lr": 0.0003, "grad_norm": 0.132402761534077, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24353532494, "tokens_per_second": 150893.4763105203, "data_wait_s": 8.761000572121702e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 166500, "loss": 0.5907998792827129, "lr": 0.0003, "grad_norm": 0.15242004391998884, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24368285250, "tokens_per_second": 150293.46602333966, "data_wait_s": 1.00840006780345e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 166600, "loss": 0.6393077410757542, "lr": 0.0003, "grad_norm": 0.14912576189850185, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24382992520, "tokens_per_second": 150285.97440987016, "data_wait_s": 1.5299001461244188e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 166700, "loss": 0.6023055128753185, "lr": 0.0003, "grad_norm": 0.16025117601346572, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24397533861, "tokens_per_second": 147846.85166529688, "data_wait_s": 1.3814000340062194e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ca500f3c1a27da27c6b15db9e51ca010ebdd1a4946afcf9feebb5a486025b264
-size 247419
+oid sha256:ff587b48474cec458786ccc7b881c8e6a562314779f98a51f40131a5b5843c4b
+size 249537