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": 75885.53888267641,
-      "required_pages_per_second_upper_estimate": 0.006577566180210461
+      "required_token_positions_per_second": 76022.21408185888,
+      "required_pages_per_second_upper_estimate": 0.006589412840075472
     },
     {
       "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": 9485.692360334551,
-      "required_pages_per_second_upper_estimate": 0.0037839845062767477
+      "required_token_positions_per_second": 9502.77676023236,
+      "required_pages_per_second_upper_estimate": 0.003790799728830525
     },
     {
       "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": 18971.384720669103,
-      "required_pages_per_second_upper_estimate": 0.004388695970159301
+      "required_token_positions_per_second": 19005.55352046472,
+      "required_pages_per_second_upper_estimate": 0.0043966003206416795
     },
     {
       "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": 28457.077081003652,
-      "required_pages_per_second_upper_estimate": 0.0017328866914884025
+      "required_token_positions_per_second": 28508.330280697082,
+      "required_pages_per_second_upper_estimate": 0.0017360077424449756
     },
     {
       "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": 28457.077081003652,
-      "required_pages_per_second_upper_estimate": 0.0021907045679267967
+      "required_token_positions_per_second": 28508.330280697082,
+      "required_pages_per_second_upper_estimate": 0.002194650181117134
     },
     {
       "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": 28457.077081003652,
-      "required_pages_per_second_upper_estimate": 0.3533416576356662
+      "required_token_positions_per_second": 28508.330280697082,
+      "required_pages_per_second_upper_estimate": 0.35397805084243367
     }
   ],
   "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": 0.07503516036721791,
-    "producer_p95_seconds": 0.16936859299994467,
-    "gpu_update_seconds": 0.971568510753912,
-    "producer_batches_per_second": 13.327085530384096,
-    "gpu_batches_per_second": 1.0292634939599121,
-    "producer_headroom": 12.948176641445286,
+    "producer_batch_seconds": 0.2869732708672039,
+    "producer_p95_seconds": 0.18901343699963036,
+    "gpu_update_seconds": 0.969821793411745,
+    "producer_batches_per_second": 3.4846450924788295,
+    "gpu_batches_per_second": 1.0311172699904905,
+    "producer_headroom": 3.3794847529912544,
     "sustainable": true,
-    "buffer_seconds": 60.23724766674255,
+    "buffer_seconds": 60.12895119152819,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 528274.2857321614,
-    "upload_bytes_per_second": 4564126.407315318,
+    "generation_bytes_per_second": 473159.25148700457,
+    "upload_bytes_per_second": 4597171.145710632,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 113.35536412213928,
+    "estimated_full_shards_per_hour": 101.52896078546145,
     "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": 5874473,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.191119417000209,
+      "bytes": 6703440,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -32,3 +32,4 @@
 {"step": 240300, "loss": 0.6386078968644142, "lr": 0.0003, "grad_norm": 0.14710874828634884, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35172619114, "tokens_per_second": 150686.73625573123, "data_wait_s": 1.1144999916723464e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 240400, "loss": 0.5970120839774609, "lr": 0.0003, "grad_norm": 0.15777480089589546, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35187296028, "tokens_per_second": 149482.5011255308, "data_wait_s": 1.2068999694747617e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 240500, "loss": 0.5818395670503378, "lr": 0.0003, "grad_norm": 0.15575061846174584, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35201777894, "tokens_per_second": 147240.0394176374, "data_wait_s": 1.1921999885089463e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 240600, "loss": 0.5745463036000729, "lr": 0.0003, "grad_norm": 0.13907632765030298, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35216349372, "tokens_per_second": 147902.87925055713, "data_wait_s": 1.183400036097737e-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 sha256:671f57c99f5445bd95093e1fabad9b682d30ef15c2de1cf8d13fe7672e3eacab
-size 72960
+oid sha256:b600514739136da2fcf5eb623f421c257b5fb516d12f3e8e5965ec11dbba81ec
+size 75078