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": 75909.1737757618,
-      "required_pages_per_second_upper_estimate": 0.0065796147928410135
+      "required_token_positions_per_second": 75821.71271923695,
+      "required_pages_per_second_upper_estimate": 0.006572033890129461
     },
     {
       "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": 9488.646721970224,
-      "required_pages_per_second_upper_estimate": 0.0037851630453048604
+      "required_token_positions_per_second": 9477.714089904619,
+      "required_pages_per_second_upper_estimate": 0.0037808018549164747
     },
     {
       "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": 18977.29344394045,
-      "required_pages_per_second_upper_estimate": 0.004390062849298148
+      "required_token_positions_per_second": 18955.428179809238,
+      "required_pages_per_second_upper_estimate": 0.0043850047052568985
     },
     {
       "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": 28465.940165910673,
-      "required_pages_per_second_upper_estimate": 0.0017334264068582274
+      "required_token_positions_per_second": 28433.142269713855,
+      "required_pages_per_second_upper_estimate": 0.0017314291870571036
     },
     {
       "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": 28465.940165910673,
-      "required_pages_per_second_upper_estimate": 0.002191386872737528
+      "required_token_positions_per_second": 28433.142269713855,
+      "required_pages_per_second_upper_estimate": 0.0021888619998979095
     },
     {
       "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": 28465.940165910673,
-      "required_pages_per_second_upper_estimate": 0.3534517074873744
+      "required_token_positions_per_second": 28433.142269713855,
+      "required_pages_per_second_upper_estimate": 0.3530444673841074
     }
   ],
   "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": 0.08727420951564113,
-    "producer_p95_seconds": 0.2581432000006316,
-    "gpu_update_seconds": 0.97126600557918,
-    "producer_batches_per_second": 11.458138727922615,
-    "gpu_batches_per_second": 1.029584062713783,
-    "producer_headroom": 11.128900633641504,
+    "producer_batch_seconds": 0.1256954141876463,
+    "producer_p95_seconds": 0.253937464000046,
+    "gpu_update_seconds": 0.9723863700231906,
+    "producer_batches_per_second": 7.955739725771816,
+    "gpu_batches_per_second": 1.0283977962136088,
+    "producer_headroom": 7.73605287279255,
     "sustainable": true,
-    "buffer_seconds": 60.218492345909155,
+    "buffer_seconds": 60.28795494143782,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 256525.05875443268,
+    "generation_bytes_per_second": 301891.4792927659,
     "upload_bytes_per_second": 5238079.44633967,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 55.04430601095901,
+    "estimated_full_shards_per_hour": 64.77888378226503,
     "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": [

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

@@ -198,3 +198,4 @@
 {"step": 256900, "loss": 0.5757004469633102, "lr": 0.0003, "grad_norm": 0.15516702809483268, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37602925941, "tokens_per_second": 147485.50615365646, "data_wait_s": 1.4310997357824817e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 257000, "loss": 0.6060768719762564, "lr": 0.0003, "grad_norm": 0.17568757799964974, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37617394237, "tokens_per_second": 146889.6447210031, "data_wait_s": 1.2466000043787062e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 257100, "loss": 0.5988826267421246, "lr": 0.0003, "grad_norm": 0.15889377992668277, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37632052782, "tokens_per_second": 149741.4795628632, "data_wait_s": 1.4068999007577077e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 257200, "loss": 0.613028310239315, "lr": 0.0003, "grad_norm": 0.14739748755808474, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37646738818, "tokens_per_second": 149730.20846703582, "data_wait_s": 1.3818000297760591e-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 sha256:3956e8302c3b26238d74c9b386d2fd7599e68df4ff9fddfcbfb446ba88ac5d36
-size 426897
+oid sha256:294ac86e49aa96afbcc75c2492eeadc050505c34d7c581c888158cf958acbc72
+size 429015