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": 75081.84257396919,
-      "required_pages_per_second_upper_estimate": 0.00650790382112139
+      "required_token_positions_per_second": 75971.8592555506,
+      "required_pages_per_second_upper_estimate": 0.006585048211354211
     },
     {
       "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": 9385.230321746149,
-      "required_pages_per_second_upper_estimate": 0.003743908697042504
+      "required_token_positions_per_second": 9496.482406943825,
+      "required_pages_per_second_upper_estimate": 0.0037882888171947605
     },
     {
       "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": 18770.460643492297,
-      "required_pages_per_second_upper_estimate": 0.004342215721047431
+      "required_token_positions_per_second": 18992.96481388765,
+      "required_pages_per_second_upper_estimate": 0.004393688144928743
     },
     {
       "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": 28155.690965238446,
-      "required_pages_per_second_upper_estimate": 0.0017145338582890454
+      "required_token_positions_per_second": 28489.447220831476,
+      "required_pages_per_second_upper_estimate": 0.0017348578631709187
     },
     {
       "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": 28155.690965238446,
-      "required_pages_per_second_upper_estimate": 0.0021675030304450253
+      "required_token_positions_per_second": 28489.447220831476,
+      "required_pages_per_second_upper_estimate": 0.0021931965108970245
     },
     {
       "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": 28155.690965238446,
-      "required_pages_per_second_upper_estimate": 0.3495994507523057
+      "required_token_positions_per_second": 28489.447220831476,
+      "required_pages_per_second_upper_estimate": 0.3537435864364389
     }
   ],
   "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": 0.08100452710948502,
-    "producer_p95_seconds": 0.21077838299970608,
-    "gpu_update_seconds": 0.9819684423349705,
-    "producer_batches_per_second": 12.34498904793813,
-    "gpu_batches_per_second": 1.0183626651200248,
-    "producer_headroom": 12.122389666046075,
+    "producer_batch_seconds": 0.0968696881327844,
+    "producer_p95_seconds": 0.20495424299951992,
+    "gpu_update_seconds": 0.9704645999513739,
+    "producer_batches_per_second": 10.323146685774885,
+    "gpu_batches_per_second": 1.0304342889478977,
+    "producer_headroom": 10.018248418649875,
     "sustainable": true,
-    "buffer_seconds": 60.88204342476817,
+    "buffer_seconds": 60.16880519698518,
     "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": 466394.9751915541,
-    "upload_bytes_per_second": 5395414.605614731,
+    "generation_bytes_per_second": 463692.299740859,
+    "upload_bytes_per_second": 5357066.625336519,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 100.07750455675094,
+    "estimated_full_shards_per_hour": 99.49757332009628,
     "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": 1.1050905230003991,
-      "bytes": 6187776,
-      "pages": 15,
-      "shards": 2
-    },
     {
       "seconds": 1.2981188600006135,
       "bytes": 10611352,
@@ -562,6 +556,12 @@
       "bytes": 6128160,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.9900568350021786,
+      "bytes": 2575859,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -103,3 +103,4 @@
 {"step": 290500, "loss": 0.5722408331930637, "lr": 0.0003, "grad_norm": 0.1425618393093551, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42520974684, "tokens_per_second": 149176.61279625515, "data_wait_s": 9.75399962044321e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 290600, "loss": 0.5969054363667965, "lr": 0.0003, "grad_norm": 0.17551084322181382, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42535497260, "tokens_per_second": 149263.4370258846, "data_wait_s": 1.0021001799032092e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 290700, "loss": 0.623919390141964, "lr": 0.0003, "grad_norm": 0.14654368036302592, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42550246399, "tokens_per_second": 151182.37926973036, "data_wait_s": 1.2541000614874065e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 290800, "loss": 0.6278765704482794, "lr": 0.0003, "grad_norm": 0.1561798517446671, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42564830823, "tokens_per_second": 150195.36210640217, "data_wait_s": 2.0540999685181305e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:711b65d734a1e8daa09b5a789b335e8e4741cf1e8de6194f79509c0a74a3fd63
-size 224121
+oid sha256:caadcdb381c95136276023cf215f05d8aafca94414879ce2fd6dc225d7460b90
+size 226239