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": 75157.22568400606,
-      "required_pages_per_second_upper_estimate": 0.006514437838042641
+      "required_token_positions_per_second": 75830.60994071847,
+      "required_pages_per_second_upper_estimate": 0.006572805078737677
     },
     {
       "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": 9394.653210500757,
-      "required_pages_per_second_upper_estimate": 0.0037476676282514588
+      "required_token_positions_per_second": 9478.826242589808,
+      "required_pages_per_second_upper_estimate": 0.0037812455092507613
     },
     {
       "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": 18789.306421001515,
-      "required_pages_per_second_upper_estimate": 0.004346575359999832
+      "required_token_positions_per_second": 18957.652485179617,
+      "required_pages_per_second_upper_estimate": 0.004385519259157898
     },
     {
       "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": 28183.95963150227,
-      "required_pages_per_second_upper_estimate": 0.0017162552717502827
+      "required_token_positions_per_second": 28436.478727769423,
+      "required_pages_per_second_upper_estimate": 0.0017316323598476453
     },
     {
       "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": 28183.95963150227,
-      "required_pages_per_second_upper_estimate": 0.0021696792306266912
+      "required_token_positions_per_second": 28436.478727769423,
+      "required_pages_per_second_upper_estimate": 0.0021891188496749345
     },
     {
       "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": 28183.95963150227,
-      "required_pages_per_second_upper_estimate": 0.34995045297816246
+      "required_token_positions_per_second": 28436.478727769423,
+      "required_pages_per_second_upper_estimate": 0.35308589502675075
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.11783737476574174,
-    "producer_p95_seconds": 0.3697225180003443,
-    "gpu_update_seconds": 0.9809835225954834,
-    "producer_batches_per_second": 8.486271880954403,
-    "gpu_batches_per_second": 1.0193851139866272,
-    "producer_headroom": 8.324892883481649,
+    "producer_batch_seconds": 0.09082973065630995,
+    "producer_p95_seconds": 0.37095496599795297,
+    "gpu_update_seconds": 0.9722722797250054,
+    "producer_batches_per_second": 11.009610980614857,
+    "gpu_batches_per_second": 1.0285184725032344,
+    "producer_headroom": 10.70433956700786,
     "sustainable": true,
-    "buffer_seconds": 60.82097840091997,
+    "buffer_seconds": 60.28088134295034,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 321072.0685865321,
-    "upload_bytes_per_second": 4134798.041822171,
+    "generation_bytes_per_second": 331943.79149414686,
+    "upload_bytes_per_second": 4185164.5493375333,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.89459174344036,
+    "estimated_full_shards_per_hour": 71.22741039865784,
     "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": 1.0474640769971302,
-      "bytes": 900238,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.1159279130006325,
       "bytes": 6352582,
@@ -562,6 +556,12 @@
       "bytes": 2671135,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.0467685680014256,
+      "bytes": 5999640,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -293,3 +293,4 @@
 {"step": 344300, "loss": 0.6537228301167488, "lr": 0.0003, "grad_norm": 0.18833676195820212, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50395982680, "tokens_per_second": 149616.36619675183, "data_wait_s": 1.4023000403540209e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 344400, "loss": 0.5523795243352652, "lr": 0.0003, "grad_norm": 0.15344267455651656, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50410762680, "tokens_per_second": 149709.6909339595, "data_wait_s": 1.4655997802037746e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 344500, "loss": 0.5823973026126623, "lr": 0.0003, "grad_norm": 0.16255057540814566, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50425496297, "tokens_per_second": 150492.55814300498, "data_wait_s": 1.4551998901879415e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 344600, "loss": 0.5281241871416569, "lr": 0.0003, "grad_norm": 0.17750216107181283, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50440090874, "tokens_per_second": 123360.64090300149, "data_wait_s": 1.1546999303391203e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6ce0fd817eb5632158e4d4ded7a2fc4b8a8c720423c21f0775db7db7aaf85a01
-size 628890
+oid sha256:863959eed2b53b228272812327de6778b4a434751b5735d439256364fefaa9fe
+size 631008