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": 75645.14073930797,
-      "required_pages_per_second_upper_estimate": 0.006556729078426808
+      "required_token_positions_per_second": 75400.95543232367,
+      "required_pages_per_second_upper_estimate": 0.006535563714899412
     },
     {
       "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": 9455.642592413496,
-      "required_pages_per_second_upper_estimate": 0.003771997204568971
+      "required_token_positions_per_second": 9425.119429040458,
+      "required_pages_per_second_upper_estimate": 0.0037598210583375054
     },
     {
       "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": 18911.285184826993,
-      "required_pages_per_second_upper_estimate": 0.004374793000257934
+      "required_token_positions_per_second": 18850.238858080917,
+      "required_pages_per_second_upper_estimate": 0.0043606710069438856
     },
     {
       "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": 28366.927777240493,
-      "required_pages_per_second_upper_estimate": 0.0017273970718660694
+      "required_token_positions_per_second": 28275.358287121373,
+      "required_pages_per_second_upper_estimate": 0.0017218209703458002
     },
     {
       "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": 28366.927777240493,
-      "required_pages_per_second_upper_estimate": 0.002183764624973864
+      "required_token_positions_per_second": 28275.358287121373,
+      "required_pages_per_second_upper_estimate": 0.0021767153521439217
     },
     {
       "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": 28366.927777240493,
-      "required_pages_per_second_upper_estimate": 0.35222230499323903
+      "required_token_positions_per_second": 28275.358287121373,
+      "required_pages_per_second_upper_estimate": 0.35108531838932877
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.13250462485922299,
-    "producer_p95_seconds": 0.35408145399560453,
-    "gpu_update_seconds": 0.9746561283306363,
-    "producer_batches_per_second": 7.546906389587766,
-    "gpu_batches_per_second": 1.026002885461534,
-    "producer_headroom": 7.355638562549354,
+    "producer_batch_seconds": 0.10460434400010854,
+    "producer_p95_seconds": 0.3212562200060347,
+    "gpu_update_seconds": 0.9778125433194911,
+    "producer_batches_per_second": 9.559832429128969,
+    "gpu_batches_per_second": 1.0226909102691468,
+    "producer_headroom": 9.347724061234747,
     "sustainable": true,
-    "buffer_seconds": 60.42867995649945,
+    "buffer_seconds": 60.62437768580845,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 342498.36349467485,
-    "upload_bytes_per_second": 4107657.7498192834,
+    "generation_bytes_per_second": 356392.9783265349,
+    "upload_bytes_per_second": 4068240.5902861375,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.49217585210975,
+    "estimated_full_shards_per_hour": 76.47363674494777,
     "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": 1.080213831999572,
-      "bytes": 6038268,
-      "pages": 31,
-      "shards": 2
-    },
     {
       "seconds": 0.7658507999985886,
       "bytes": 1224241,
@@ -562,6 +556,12 @@
       "bytes": 7491850,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 0.8147378479989129,
+      "bytes": 902236,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -256,3 +256,4 @@
 {"step": 110600, "loss": 0.6474052537232637, "lr": 0.0003, "grad_norm": 0.13868067379159402, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16184359123, "tokens_per_second": 149237.21902385305, "data_wait_s": 1.58320035552606e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 110700, "loss": 0.6160413213074207, "lr": 0.0003, "grad_norm": 0.12751857205388983, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16198989215, "tokens_per_second": 149054.01990013264, "data_wait_s": 1.3787997886538506e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 110800, "loss": 0.6187618598341942, "lr": 0.0003, "grad_norm": 0.14943964565670242, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16213455540, "tokens_per_second": 147097.4556200023, "data_wait_s": 1.4114993973635137e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 110900, "loss": 0.6345473565161228, "lr": 0.0003, "grad_norm": 0.13995303506369336, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16227984156, "tokens_per_second": 147334.913506741, "data_wait_s": 1.5177000022958964e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:93ccfa4337ac2e513ed4e4bbfb1897de68d4ca882d424740d39b44c02b7fbb01
-size 550524
+oid sha256:ba931ff28c6381c5ecbece3118db7f931e749b63d9521007f5343f24b07eaab7
+size 552642