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": 75644.91298961571,
-      "required_pages_per_second_upper_estimate": 0.006556709337660709
+      "required_token_positions_per_second": 75041.13704577973,
+      "required_pages_per_second_upper_estimate": 0.006504375569105146
     },
     {
       "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": 9455.614123701964,
-      "required_pages_per_second_upper_estimate": 0.003771985847974295
+      "required_token_positions_per_second": 9380.142130722466,
+      "required_pages_per_second_upper_estimate": 0.003741878941567922
     },
     {
       "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": 18911.228247403928,
-      "required_pages_per_second_upper_estimate": 0.004374779828787172
+      "required_token_positions_per_second": 18760.284261444933,
+      "required_pages_per_second_upper_estimate": 0.00433986159415898
     },
     {
       "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": 28366.84237110589,
-      "required_pages_per_second_upper_estimate": 0.0017273918710805634
+      "required_token_positions_per_second": 28140.4263921674,
+      "required_pages_per_second_upper_estimate": 0.0017136043258760441
     },
     {
       "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": 28366.84237110589,
-      "required_pages_per_second_upper_estimate": 0.002183758050173203
+      "required_token_positions_per_second": 28140.4263921674,
+      "required_pages_per_second_upper_estimate": 0.0021663279213549737
     },
     {
       "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": 28366.84237110589,
-      "required_pages_per_second_upper_estimate": 0.35222124453488324
+      "required_token_positions_per_second": 28140.4263921674,
+      "required_pages_per_second_upper_estimate": 0.34940991584200304
     }
   ],
   "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": 0.14419395472634733,
-    "producer_p95_seconds": 0.4013354740018258,
-    "gpu_update_seconds": 0.9746590627993867,
-    "producer_batches_per_second": 6.935103499295859,
-    "gpu_batches_per_second": 1.0259997964086331,
-    "producer_headroom": 6.759361477040449,
+    "producer_batch_seconds": 0.3106612933511883,
+    "producer_p95_seconds": 0.38700950700149406,
+    "gpu_update_seconds": 0.9825011040946963,
+    "producer_batches_per_second": 3.2189397952114556,
+    "gpu_batches_per_second": 1.0178105610592953,
+    "producer_headroom": 3.1626119028096107,
     "sustainable": true,
-    "buffer_seconds": 60.42886189356198,
+    "buffer_seconds": 60.91506845387117,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 473935.45950239187,
-    "upload_bytes_per_second": 5649704.800914396,
+    "generation_bytes_per_second": 456151.7154215424,
+    "upload_bytes_per_second": 5690215.16228837,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.69551695636575,
+    "estimated_full_shards_per_hour": 97.87953946098999,
     "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": 0.8884790899974178,
-      "bytes": 1189196,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.1859263639998971,
       "bytes": 7316047,
@@ -562,6 +556,12 @@
       "bytes": 6123455,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.302649115998065,
+      "bytes": 6509625,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -350,3 +350,4 @@
 {"step": 230000, "loss": 0.6125554777681828, "lr": 0.0003, "grad_norm": 0.14557966935096453, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33665466237, "tokens_per_second": 149953.93094908562, "data_wait_s": 1.871899439720437e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 230100, "loss": 0.6135698668658733, "lr": 0.0003, "grad_norm": 0.15768517799055218, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33680208565, "tokens_per_second": 131540.78586851037, "data_wait_s": 1.205400621984154e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 230200, "loss": 0.6240835413336754, "lr": 0.0003, "grad_norm": 0.14319430610520004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33695018223, "tokens_per_second": 149315.89953658028, "data_wait_s": 1.1395000910852104e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 230300, "loss": 0.6145809032022953, "lr": 0.0003, "grad_norm": 0.14974299864634225, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33709652932, "tokens_per_second": 148762.6084711953, "data_wait_s": 1.1199001164641231e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4c9e1bc3f91308bff15eb07036af714bddfcf61c627a4e24fd1aac1d93459938
-size 751182
+oid sha256:25b99294a933555604a90e8e402f3967af4e1dbf2b45f5e513247c3cab15b3df
+size 753300