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": 75063.6696982732,
-      "required_pages_per_second_upper_estimate": 0.006506328642314794
+      "required_token_positions_per_second": 75236.87356143583,
+      "required_pages_per_second_upper_estimate": 0.006521341514192553
     },
     {
       "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": 9382.95871228415,
-      "required_pages_per_second_upper_estimate": 0.0037430025180645245
+      "required_token_positions_per_second": 9404.609195179479,
+      "required_pages_per_second_upper_estimate": 0.003751639219395037
     },
     {
       "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": 18765.9174245683,
-      "required_pages_per_second_upper_estimate": 0.004341164727307286
+      "required_token_positions_per_second": 18809.218390358958,
+      "required_pages_per_second_upper_estimate": 0.004351181643671985
     },
     {
       "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": 28148.876136852447,
-      "required_pages_per_second_upper_estimate": 0.0017141188709949792
+      "required_token_positions_per_second": 28213.827585538435,
+      "required_pages_per_second_upper_estimate": 0.0017180740734460411
     },
     {
       "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": 28148.876136852447,
-      "required_pages_per_second_upper_estimate": 0.0021669784060912188
+      "required_token_positions_per_second": 28213.827585538435,
+      "required_pages_per_second_upper_estimate": 0.0021719785484081836
     },
     {
       "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": 28148.876136852447,
-      "required_pages_per_second_upper_estimate": 0.349514833391515
+      "required_token_positions_per_second": 28213.827585538435,
+      "required_pages_per_second_upper_estimate": 0.3503213130056798
     }
   ],
   "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": 0.119065797507659,
-    "producer_p95_seconds": 0.23404077099985443,
-    "gpu_update_seconds": 0.9822061763880974,
-    "producer_batches_per_second": 8.398717523692513,
-    "gpu_batches_per_second": 1.0181161797183322,
-    "producer_headroom": 8.249272225509733,
+    "producer_batch_seconds": 0.26326494410929513,
+    "producer_p95_seconds": 0.22448090900070383,
+    "gpu_update_seconds": 0.9799450257565031,
+    "producer_batches_per_second": 3.7984548356155137,
+    "gpu_batches_per_second": 1.020465407463051,
+    "producer_headroom": 3.722276921722158,
     "sustainable": true,
-    "buffer_seconds": 60.89678293606204,
+    "buffer_seconds": 60.75659159690319,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 337505.1929267124,
-    "upload_bytes_per_second": 4806990.720324847,
+    "generation_bytes_per_second": 328564.70866217755,
+    "upload_bytes_per_second": 4859561.811817777,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.42075768328694,
+    "estimated_full_shards_per_hour": 70.50233788393969,
     "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": 1.197562700999697,
-      "bytes": 7759288,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 0.8769548000000214,
       "bytes": 1135557,
@@ -562,6 +556,12 @@
       "bytes": 4762409,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.406141793999268,
+      "bytes": 13345421,
+      "pages": 38,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -171,3 +171,4 @@
 {"step": 254200, "loss": 0.632264755666256, "lr": 0.0003, "grad_norm": 0.13589899055627355, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37208166907, "tokens_per_second": 149553.803815343, "data_wait_s": 1.0056999599328265e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 254300, "loss": 0.5514777507632971, "lr": 0.0003, "grad_norm": 0.1366859609829462, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37222800816, "tokens_per_second": 135755.4666257701, "data_wait_s": 9.111001418204978e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 254400, "loss": 0.5483874548226595, "lr": 0.0003, "grad_norm": 0.14924686155782332, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37237388209, "tokens_per_second": 149804.04372506248, "data_wait_s": 7.629001629538834e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 254500, "loss": 0.553221432492137, "lr": 0.0003, "grad_norm": 0.14108443836552215, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37251860553, "tokens_per_second": 147419.75755384032, "data_wait_s": 1.4804001693846658e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3154fbfbe8ddeff529945fe2c5db8a214b7cda6589d52eadd975346642b8695d
-size 368928
+oid sha256:251b503e91006a4017844ed09f4bd86d49ef42d2ff8ded51ad275e8634a71225
+size 371046