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": 75522.05666764315,
-      "required_pages_per_second_upper_estimate": 0.006546060463048629
+      "required_token_positions_per_second": 75608.98886691632,
+      "required_pages_per_second_upper_estimate": 0.006553595525753983
     },
     {
       "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": 9440.257083455394,
-      "required_pages_per_second_upper_estimate": 0.0037658596950117256
+      "required_token_positions_per_second": 9451.12360836454,
+      "required_pages_per_second_upper_estimate": 0.0037701945142670098
     },
     {
       "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": 18880.514166910787,
-      "required_pages_per_second_upper_estimate": 0.004367674666814439
+      "required_token_positions_per_second": 18902.24721672908,
+      "required_pages_per_second_upper_estimate": 0.004372702225401288
     },
     {
       "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": 28320.77125036618,
-      "required_pages_per_second_upper_estimate": 0.0017245863815440047
+      "required_token_positions_per_second": 28353.370825093618,
+      "required_pages_per_second_upper_estimate": 0.0017265715246081556
     },
     {
       "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": 28320.77125036618,
-      "required_pages_per_second_upper_estimate": 0.0021802113677656353
+      "required_token_positions_per_second": 28353.370825093618,
+      "required_pages_per_second_upper_estimate": 0.0021827209732695517
     },
     {
       "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": 28320.77125036618,
-      "required_pages_per_second_upper_estimate": 0.35164919540541834
+      "required_token_positions_per_second": 28353.370825093618,
+      "required_pages_per_second_upper_estimate": 0.352053973019775
     }
   ],
   "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": 0.10973597222658782,
-    "producer_p95_seconds": 0.3095788829996309,
-    "gpu_update_seconds": 0.9762445999645055,
-    "producer_batches_per_second": 9.11278206871995,
-    "gpu_batches_per_second": 1.0243334508957675,
-    "producer_headroom": 8.896304285241227,
+    "producer_batch_seconds": 0.07593554861722396,
+    "producer_p95_seconds": 0.3145949589998054,
+    "gpu_update_seconds": 0.9751221528669408,
+    "producer_batches_per_second": 13.169062688159425,
+    "gpu_batches_per_second": 1.0255125443103885,
+    "producer_headroom": 12.84144475971772,
     "sustainable": true,
-    "buffer_seconds": 60.52716519779934,
+    "buffer_seconds": 60.45757347775033,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 329475.0214938009,
-    "upload_bytes_per_second": 6076827.066240456,
+    "generation_bytes_per_second": 316202.6023194876,
+    "upload_bytes_per_second": 6056738.078214915,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.6976698266079,
+    "estimated_full_shards_per_hour": 67.8497176379058,
     "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": 1.1391355480009224,
-      "bytes": 7817543,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.0901677820002078,
       "bytes": 6115749,
@@ -562,6 +556,12 @@
       "bytes": 8367852,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.080213831999572,
+      "bytes": 6038268,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -178,3 +178,4 @@
 {"step": 102800, "loss": 0.587541026994586, "lr": 0.0003, "grad_norm": 0.1717889755210538, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15043150245, "tokens_per_second": 148986.56333670835, "data_wait_s": 1.480399805586785e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 102900, "loss": 0.5999361556023359, "lr": 0.0003, "grad_norm": 0.13805758448796915, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15057865595, "tokens_per_second": 148841.67114829342, "data_wait_s": 1.5591998817399144e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 103000, "loss": 0.638515941798687, "lr": 0.0003, "grad_norm": 0.13897537475620358, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15072547623, "tokens_per_second": 149250.89938169898, "data_wait_s": 1.4072000340092927e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 103100, "loss": 0.6555536761879921, "lr": 0.0003, "grad_norm": 0.1502817850151909, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15087338923, "tokens_per_second": 150393.0267746047, "data_wait_s": 1.6738998965593055e-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 sha256:605d0da4ccef8c19b3113894d73aaa27cd5d72b8ece14556c4a37c9defe76e4f
-size 383754
+oid sha256:e28555b2a86e78048c178f59d9402a8d5a1dd0495a4c2dd428726355f7e18150
+size 385872