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": 75311.6631111373,
-      "required_pages_per_second_upper_estimate": 0.0065278240828082965
+      "required_token_positions_per_second": 75378.38385708103,
+      "required_pages_per_second_upper_estimate": 0.006533607267964515
     },
     {
       "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": 9413.957888892162,
-      "required_pages_per_second_upper_estimate": 0.0037553685530924536
+      "required_token_positions_per_second": 9422.29798213513,
+      "required_pages_per_second_upper_estimate": 0.003758695540982579
     },
     {
       "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": 18827.915777784325,
-      "required_pages_per_second_upper_estimate": 0.004355506955190064
+      "required_token_positions_per_second": 18844.59596427026,
+      "required_pages_per_second_upper_estimate": 0.004359365622772329
     },
     {
       "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": 28241.87366667649,
-      "required_pages_per_second_upper_estimate": 0.0017197819326409329
+      "required_token_positions_per_second": 28266.893946405387,
+      "required_pages_per_second_upper_estimate": 0.0017213055364051603
     },
     {
       "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": 28241.87366667649,
-      "required_pages_per_second_upper_estimate": 0.0021741376133707128
+      "required_token_positions_per_second": 28266.893946405387,
+      "required_pages_per_second_upper_estimate": 0.0021760637437648104
     },
     {
       "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": 28241.87366667649,
-      "required_pages_per_second_upper_estimate": 0.35066955146921897
+      "required_token_positions_per_second": 28266.893946405387,
+      "required_pages_per_second_upper_estimate": 0.3509802196059623
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.43002914478904586,
-    "producer_p95_seconds": 0.211111577000338,
-    "gpu_update_seconds": 0.9789718743987861,
-    "producer_batches_per_second": 2.325423781428949,
-    "gpu_batches_per_second": 1.0214798056523615,
-    "producer_headroom": 2.276524478077011,
+    "producer_batch_seconds": 0.07178548859372569,
+    "producer_p95_seconds": 0.19139098899995588,
+    "gpu_update_seconds": 0.9781053430356083,
+    "producer_batches_per_second": 13.930392055413323,
+    "gpu_batches_per_second": 1.0223847636865375,
+    "producer_headroom": 13.625390899980559,
     "sustainable": true,
-    "buffer_seconds": 60.696256212724734,
+    "buffer_seconds": 60.64253126820771,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 332681.8403673356,
-    "upload_bytes_per_second": 3415736.8561639646,
+    "generation_bytes_per_second": 350385.83107122744,
+    "upload_bytes_per_second": 3439778.063349142,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.3857785059457,
+    "estimated_full_shards_per_hour": 75.18464278318994,
     "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": [
@@ -490,6 +490,12 @@
       "bytes": 7487764,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.1006291130006502,
+      "bytes": 6137509,
+      "pages": 39,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -62,3 +62,4 @@
 {"step": 243300, "loss": 0.591025285422802, "lr": 0.0003, "grad_norm": 0.1752452655497461, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35611473614, "tokens_per_second": 147379.8531822967, "data_wait_s": 1.3977999515191186e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 243400, "loss": 0.5518008135259151, "lr": 0.0003, "grad_norm": 0.1428783108965096, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35626103164, "tokens_per_second": 147795.38314583182, "data_wait_s": 1.4405000001715962e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 243500, "loss": 0.5820450047031045, "lr": 0.0003, "grad_norm": 0.146134232664598, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35640810545, "tokens_per_second": 150072.78225572026, "data_wait_s": 9.190000128000975e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 243600, "loss": 0.5845957063138485, "lr": 0.0003, "grad_norm": 0.15045103453201752, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35655504986, "tokens_per_second": 149962.21649101932, "data_wait_s": 3.174199991917703e-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:e9d8d0cf3e800e3c20e813b6193c870641626d8625a6b8f526b63254c1610560
-size 136500
+oid sha256:e2564fc25fd37b844e9923a415d1d98d775b0c9263bb1b4b0bee3055eeb39238
+size 138618