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": 74466.5505571033,
-      "required_pages_per_second_upper_estimate": 0.006454571868542815
+      "required_token_positions_per_second": 74616.67922449438,
+      "required_pages_per_second_upper_estimate": 0.006467584640934645
     },
     {
       "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": 9308.318819637912,
-      "required_pages_per_second_upper_estimate": 0.003713227548922097
+      "required_token_positions_per_second": 9327.084903061797,
+      "required_pages_per_second_upper_estimate": 0.0037207136201778354
     },
     {
       "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": 18616.637639275825,
-      "required_pages_per_second_upper_estimate": 0.004306631476214362
+      "required_token_positions_per_second": 18654.169806123595,
+      "required_pages_per_second_upper_estimate": 0.004315313882471018
     },
     {
       "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": 27924.95645891374,
-      "required_pages_per_second_upper_estimate": 0.0017004833374242676
+      "required_token_positions_per_second": 27981.25470918539,
+      "required_pages_per_second_upper_estimate": 0.0017039116054917195
     },
     {
       "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": 27924.95645891374,
-      "required_pages_per_second_upper_estimate": 0.002149740449434157
+      "required_token_positions_per_second": 27981.25470918539,
+      "required_pages_per_second_upper_estimate": 0.0021540744445835993
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27924.95645891374,
-      "required_pages_per_second_upper_estimate": 0.34673450040247017
+      "required_token_positions_per_second": 27981.25470918539,
+      "required_pages_per_second_upper_estimate": 0.34743353625272094
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07753124126560351,
-    "producer_p95_seconds": 0.21058728999923915,
-    "gpu_update_seconds": 0.990082116714444,
-    "producer_batches_per_second": 12.898026442969472,
-    "gpu_batches_per_second": 1.0100172330336277,
-    "producer_headroom": 12.770105322094086,
+    "producer_batch_seconds": 0.4054135304687634,
+    "producer_p95_seconds": 0.24095169599968358,
+    "gpu_update_seconds": 0.9880900727058536,
+    "producer_batches_per_second": 2.466617231160835,
+    "gpu_batches_per_second": 1.012053483405143,
+    "producer_headroom": 2.4372399992752207,
     "sustainable": true,
-    "buffer_seconds": 60.39500911958108,
+    "buffer_seconds": 60.27349443505707,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 360190.82006140036,
-    "upload_bytes_per_second": 5250738.809752594,
+    "generation_bytes_per_second": 316621.66056762595,
+    "upload_bytes_per_second": 5249086.6351809865,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.28856517201908,
+    "estimated_full_shards_per_hour": 67.93963778277954,
     "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": [
@@ -454,6 +454,12 @@
       "bytes": 13042114,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 1.1432860509994498,
+      "bytes": 5906480,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -84,3 +84,4 @@
 {"step": 203400, "loss": 0.6004175543785095, "lr": 0.0003, "grad_norm": 0.15220060382221018, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29770145221, "tokens_per_second": 148313.08790736206, "data_wait_s": 1.804100065783132e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 203500, "loss": 0.5775925777852535, "lr": 0.0003, "grad_norm": 0.1425142528755673, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29784752155, "tokens_per_second": 147981.58452242424, "data_wait_s": 1.9011000404134393e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 203600, "loss": 0.5829173810780048, "lr": 0.0003, "grad_norm": 0.15844119669134157, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29799460784, "tokens_per_second": 148556.21699054367, "data_wait_s": 1.7874001059681177e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 203700, "loss": 0.6329334229230881, "lr": 0.0003, "grad_norm": 0.1702569485568475, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29814096252, "tokens_per_second": 147765.11220061584, "data_wait_s": 1.8253998860018328e-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 sha256:1daea122f4e57dc7eefa8b07da34b88a19f6e002cfb1fa4499897991b7cecc02
-size 183096
+oid sha256:6017420b097520ca60712adbcaf4614d96a979759a58ed4ca084e7e5fb9cac7e
+size 185214