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": 76859.57643182231,
-      "required_pages_per_second_upper_estimate": 0.005317323836341079
+      "required_token_positions_per_second": 77091.17726251956,
+      "required_pages_per_second_upper_estimate": 0.00533334651919668
     },
     {
       "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": 9607.447053977789,
-      "required_pages_per_second_upper_estimate": 0.004022437455192035
+      "required_token_positions_per_second": 9636.397157814945,
+      "required_pages_per_second_upper_estimate": 0.0040345582591217385
     },
     {
       "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": 19214.894107955577,
-      "required_pages_per_second_upper_estimate": 0.004001955711853347
+      "required_token_positions_per_second": 19272.79431562989,
+      "required_pages_per_second_upper_estimate": 0.004014014798180749
     },
     {
       "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": 28822.341161933364,
-      "required_pages_per_second_upper_estimate": 0.0017278553434911643
+      "required_token_positions_per_second": 28909.191473444833,
+      "required_pages_per_second_upper_estimate": 0.0017330618870535297
     },
     {
       "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": 28822.341161933364,
-      "required_pages_per_second_upper_estimate": 0.002240810249259657
+      "required_token_positions_per_second": 28909.191473444833,
+      "required_pages_per_second_upper_estimate": 0.002247562478965526
     },
     {
       "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": 28822.341161933364,
-      "required_pages_per_second_upper_estimate": 0.3697020454064643
+      "required_token_positions_per_second": 28909.191473444833,
+      "required_pages_per_second_upper_estimate": 0.3708160679499344
     }
   ],
   "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": 0.10632929039860528,
-    "producer_p95_seconds": 0.29795005100095295,
-    "gpu_update_seconds": 0.9592558718483162,
-    "producer_batches_per_second": 9.404746295693487,
-    "gpu_batches_per_second": 1.0424747237389094,
-    "producer_headroom": 9.021558107387678,
+    "producer_batch_seconds": 0.40204866753146007,
+    "producer_p95_seconds": 0.31706355799906305,
+    "gpu_update_seconds": 0.9563740316084824,
+    "producer_batches_per_second": 2.4872610724962807,
+    "gpu_batches_per_second": 1.0456160110476285,
+    "producer_headroom": 2.378751899566106,
     "sustainable": true,
-    "buffer_seconds": 60.43311992644392,
+    "buffer_seconds": 60.25156399133439,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 330608.2101883814,
-    "upload_bytes_per_second": 5934987.030013045,
+    "generation_bytes_per_second": 322231.5663730297,
+    "upload_bytes_per_second": 5933470.289019277,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.94082574118215,
+    "estimated_full_shards_per_hour": 69.14339297669572,
     "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": 1.584766658001172,
-      "bytes": 10894182,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.1065202139980101,
       "bytes": 8432247,
@@ -562,6 +556,12 @@
       "bytes": 1265428,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.4639245990001655,
+      "bytes": 10067426,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -301,3 +301,4 @@
 {"step": 427200, "loss": 0.5943984389305115, "lr": 0.0003, "grad_norm": 0.14836201436518875, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62530168164, "tokens_per_second": 152614.19916724236, "data_wait_s": 1.2024000170640647e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 427300, "loss": 0.5756198167800903, "lr": 0.0003, "grad_norm": 0.168894812228803, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62544784988, "tokens_per_second": 152351.00901860333, "data_wait_s": 1.0269999620504677e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 427400, "loss": 0.5921719670295715, "lr": 0.0003, "grad_norm": 0.16026334633541897, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62559354053, "tokens_per_second": 151263.85183425818, "data_wait_s": 1.1219999578315765e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 427500, "loss": 0.5645163059234619, "lr": 0.0003, "grad_norm": 0.18016948607630323, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62573916972, "tokens_per_second": 150934.04924435352, "data_wait_s": 1.7066999134840444e-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 sha256:8f6d072e6ca1cf6e8ed97e7aa8a773f2fba36dd983b5d31a8d59422d3a1a4b0b
-size 646617
+oid sha256:5e2eef7da0f9a17377d4ff4c3ef9c32a22e7f13736c4cf8f5a888c49374ae5ee
+size 648735