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": 76868.59580666937,
-      "required_pages_per_second_upper_estimate": 0.005317947817620832
+      "required_token_positions_per_second": 76627.57730761699,
+      "required_pages_per_second_upper_estimate": 0.005301273598616422
     },
     {
       "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": 9608.574475833671,
-      "required_pages_per_second_upper_estimate": 0.0040229094831798476
+      "required_token_positions_per_second": 9578.447163452123,
+      "required_pages_per_second_upper_estimate": 0.004010295806615516
     },
     {
       "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": 19217.148951667343,
-      "required_pages_per_second_upper_estimate": 0.00400242533633428
+      "required_token_positions_per_second": 19156.894326904247,
+      "required_pages_per_second_upper_estimate": 0.00398987588701744
     },
     {
       "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": 28825.723427501016,
-      "required_pages_per_second_upper_estimate": 0.0017280581051475246
+      "required_token_positions_per_second": 28735.34149035637,
+      "required_pages_per_second_upper_estimate": 0.0017226398460209304
     },
     {
       "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": 28825.723427501016,
-      "required_pages_per_second_upper_estimate": 0.002241073205530528
+      "required_token_positions_per_second": 28735.34149035637,
+      "required_pages_per_second_upper_estimate": 0.0022340464074656593
     },
     {
       "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": 28825.723427501016,
-      "required_pages_per_second_upper_estimate": 0.36974542947757233
+      "required_token_positions_per_second": 28735.34149035637,
+      "required_pages_per_second_upper_estimate": 0.36858610703244404
     }
   ],
   "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": 0.33488357228139876,
-    "producer_p95_seconds": 0.24306153900033678,
-    "gpu_update_seconds": 0.9591433175835783,
-    "producer_batches_per_second": 2.9861124365924754,
-    "gpu_batches_per_second": 1.0425970568395913,
-    "producer_headroom": 2.8641097891108895,
+    "producer_batch_seconds": 0.09846661636720455,
+    "producer_p95_seconds": 0.23741445099949487,
+    "gpu_update_seconds": 0.9621601333423762,
+    "producer_batches_per_second": 10.155726244017274,
+    "gpu_batches_per_second": 1.039328034228746,
+    "producer_headroom": 9.771434917132328,
     "sustainable": true,
-    "buffer_seconds": 60.42602900776543,
+    "buffer_seconds": 60.616088400569694,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 436471.1442830071,
-    "upload_bytes_per_second": 5731018.60900511,
+    "generation_bytes_per_second": 435272.8973636265,
+    "upload_bytes_per_second": 5802953.682038003,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.65654703490885,
+    "estimated_full_shards_per_hour": 93.39943113977047,
     "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": 1.0410123799993016,
-      "bytes": 5644158,
-      "pages": 19,
-      "shards": 2
-    },
     {
       "seconds": 1.434721014000388,
       "bytes": 6921105,
@@ -562,6 +556,12 @@
       "bytes": 6769711,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.1388595060016087,
+      "bytes": 11609898,
+      "pages": 38,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -191,3 +191,4 @@
 {"step": 704100, "loss": 0.5729023814201355, "lr": 8e-05, "grad_norm": 0.21018539089421961, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103060959449, "tokens_per_second": 152121.52362965216, "data_wait_s": 1.2620999768842012e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 704200, "loss": 0.488640159368515, "lr": 8e-05, "grad_norm": 0.19922749475313972, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103075464802, "tokens_per_second": 150803.63773706026, "data_wait_s": 1.754800177877769e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 704300, "loss": 0.5114798545837402, "lr": 8e-05, "grad_norm": 0.21010549496634282, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103089904865, "tokens_per_second": 150301.74092017443, "data_wait_s": 1.6532001609448344e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 704400, "loss": 0.5500547289848328, "lr": 8e-05, "grad_norm": 0.24042668037985715, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103104545973, "tokens_per_second": 152359.51163405934, "data_wait_s": 1.5125999198062345e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a70d5c61b3f916296adfa9e165f95e551d334d6f9c0900f75a03f2b9cbcfc096
-size 411288
+oid sha256:9809409126716d84061c59252d1c5c8040ab31a8c4eec82f917de87d453549df
+size 413406