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": 75176.59749530318,
-      "required_pages_per_second_upper_estimate": 0.006516116937548468
+      "required_token_positions_per_second": 75545.13767113368,
+      "required_pages_per_second_upper_estimate": 0.00654806106593292
     },
     {
       "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": 9397.074686912898,
-      "required_pages_per_second_upper_estimate": 0.0037486335913965605
+      "required_token_positions_per_second": 9443.14220889171,
+      "required_pages_per_second_upper_estimate": 0.0037670106146847414
     },
     {
       "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": 18794.149373825796,
-      "required_pages_per_second_upper_estimate": 0.004347695691902671
+      "required_token_positions_per_second": 18886.28441778342,
+      "required_pages_per_second_upper_estimate": 0.004369009512800876
     },
     {
       "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": 28191.224060738696,
-      "required_pages_per_second_upper_estimate": 0.0017166976373772663
+      "required_token_positions_per_second": 28329.42662667513,
+      "required_pages_per_second_upper_estimate": 0.0017251134485499699
     },
     {
       "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": 28191.224060738696,
-      "required_pages_per_second_upper_estimate": 0.0021702384664985387
+      "required_token_positions_per_second": 28329.42662667513,
+      "required_pages_per_second_upper_estimate": 0.0021808776825935133
     },
     {
       "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": 28191.224060738696,
-      "required_pages_per_second_upper_estimate": 0.3500406528767982
+      "required_token_positions_per_second": 28329.42662667513,
+      "required_pages_per_second_upper_estimate": 0.3517566662114945
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.4260878572578406,
-    "producer_p95_seconds": 0.22787374499966973,
-    "gpu_update_seconds": 0.9807307387728782,
-    "producer_batches_per_second": 2.3469338141567015,
-    "gpu_batches_per_second": 1.019647860993153,
-    "producer_headroom": 2.3017101334089505,
+    "producer_batch_seconds": 0.05702117564044329,
+    "producer_p95_seconds": 0.19168829000045662,
+    "gpu_update_seconds": 0.9759463318599786,
+    "producer_batches_per_second": 17.537344482437014,
+    "gpu_batches_per_second": 1.0246465070412012,
+    "producer_headroom": 17.11550701819924,
     "sustainable": true,
-    "buffer_seconds": 60.80530580391845,
+    "buffer_seconds": 60.50867257531867,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 349299.38277791924,
-    "upload_bytes_per_second": 5937710.310092586,
+    "generation_bytes_per_second": 349113.05009194155,
+    "upload_bytes_per_second": 5813751.438737849,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.95151627066787,
+    "estimated_full_shards_per_hour": 74.91153361385999,
     "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": [
@@ -346,6 +346,12 @@
       "bytes": 13299747,
       "pages": 33,
       "shards": 4
+    },
+    {
+      "seconds": 0.8506751649983926,
+      "bytes": 1134712,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -80,3 +80,4 @@
 {"step": 93000, "loss": 0.5979013331234455, "lr": 0.0003, "grad_norm": 0.1368685890683131, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13609897800, "tokens_per_second": 149419.68909826386, "data_wait_s": 1.0979001672239974e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 93100, "loss": 0.6072111669927835, "lr": 0.0003, "grad_norm": 0.12757816160585747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13624640472, "tokens_per_second": 149375.8776522655, "data_wait_s": 1.1160998838022351e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 93200, "loss": 0.6181947682052851, "lr": 0.0003, "grad_norm": 0.15533395151872798, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13639367228, "tokens_per_second": 149947.11374554937, "data_wait_s": 1.005100057227537e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 93300, "loss": 0.5518498849123716, "lr": 0.0003, "grad_norm": 0.14331970414926384, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13653941869, "tokens_per_second": 148786.2114419632, "data_wait_s": 1.0289000783814117e-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:e37ab6f0eada3b028895da3520ba38544b2d0f9e62dbabc18499a9419d150ae2
-size 174624
+oid sha256:d955994099844165b077616388f4013e6abe6493042b69794b4f84b75debb835
+size 176742