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": 76689.49613144736,
-      "required_pages_per_second_upper_estimate": 0.005305557286520473
+      "required_token_positions_per_second": 76317.45283375047,
+      "required_pages_per_second_upper_estimate": 0.005279818467926418
     },
     {
       "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": 9586.18701643092,
-      "required_pages_per_second_upper_estimate": 0.0040135363214312296
+      "required_token_positions_per_second": 9539.68160421881,
+      "required_pages_per_second_upper_estimate": 0.003994065476481458
     },
     {
       "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": 19172.37403286184,
-      "required_pages_per_second_upper_estimate": 0.003993099901541108
+      "required_token_positions_per_second": 19079.36320843762,
+      "required_pages_per_second_upper_estimate": 0.00397372819967393
     },
     {
       "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": 28758.561049292763,
-      "required_pages_per_second_upper_estimate": 0.0017240318231249534
+      "required_token_positions_per_second": 28619.04481265643,
+      "required_pages_per_second_upper_estimate": 0.0017156680377675639
     },
     {
       "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": 28758.561049292763,
-      "required_pages_per_second_upper_estimate": 0.0022358516260409175
+      "required_token_positions_per_second": 28619.04481265643,
+      "required_pages_per_second_upper_estimate": 0.002225004852309513
     },
     {
       "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": 28758.561049292763,
-      "required_pages_per_second_upper_estimate": 0.36888394260326013
+      "required_token_positions_per_second": 28619.04481265643,
+      "required_pages_per_second_upper_estimate": 0.36709437812055296
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.2571984618047054,
+    "producer_batch_seconds": 0.09660181003880552,
     "producer_p95_seconds": 0.2634448069984501,
-    "gpu_update_seconds": 0.9613832887052576,
-    "producer_batches_per_second": 3.888048135992799,
-    "gpu_batches_per_second": 1.0401678620259247,
-    "producer_headroom": 3.7379045036251037,
+    "gpu_update_seconds": 0.9660699782605254,
+    "producer_batches_per_second": 10.35177290775705,
+    "gpu_batches_per_second": 1.035121701846659,
+    "producer_headroom": 10.00053702795475,
     "sustainable": true,
-    "buffer_seconds": 60.56714718843123,
+    "buffer_seconds": 60.8624086304131,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 405266.87990406563,
-    "upload_bytes_per_second": 4969973.832112791,
+    "generation_bytes_per_second": 404997.5381520437,
+    "upload_bytes_per_second": 4963828.883437685,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.96083829728582,
+    "estimated_full_shards_per_hour": 86.903043827257,
     "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": 1.1869826189995365,
-      "bytes": 6445004,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.1019184790002328,
       "bytes": 6210962,
@@ -562,6 +556,12 @@
       "bytes": 6431784,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.2564982580006472,
+      "bytes": 6256940,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -233,3 +233,4 @@
 {"step": 783300, "loss": 0.49220219254493713, "lr": 3e-05, "grad_norm": 0.26907296387154755, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114653339897, "tokens_per_second": 152946.0809744229, "data_wait_s": 1.1788000847445801e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 783400, "loss": 0.5688099265098572, "lr": 3e-05, "grad_norm": 0.2655296581950597, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114667796749, "tokens_per_second": 149607.83377231544, "data_wait_s": 1.9279999833088368e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 783500, "loss": 0.5408061146736145, "lr": 3e-05, "grad_norm": 0.2619653487019164, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114682268048, "tokens_per_second": 149772.36378437467, "data_wait_s": 1.0682000720407814e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 783600, "loss": 0.5225484371185303, "lr": 3e-05, "grad_norm": 0.2480367554437786, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114696800610, "tokens_per_second": 151193.368052761, "data_wait_s": 1.4572997315553948e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ab91caef79d0bdae6fad39c22599be5cde0adfc991279dc2b04b1158eea5b990
-size 501027
+oid sha256:bd5b04d61c7e952ed4b014b65c01d838d77526c60a3c8219c3ec580a59b2cfc6
+size 503145