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": 75730.10435254397,
-      "required_pages_per_second_upper_estimate": 0.006564093509083796
+      "required_token_positions_per_second": 75340.90873963531,
+      "required_pages_per_second_upper_estimate": 0.006530359019764136
     },
     {
       "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": 9466.263044067997,
-      "required_pages_per_second_upper_estimate": 0.003776233861523854
+      "required_token_positions_per_second": 9417.613592454414,
+      "required_pages_per_second_upper_estimate": 0.003756826867901075
     },
     {
       "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": 18932.526088135994,
-      "required_pages_per_second_upper_estimate": 0.004379706709411341
+      "required_token_positions_per_second": 18835.22718490883,
+      "required_pages_per_second_upper_estimate": 0.0043571983205519475
     },
     {
       "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": 28398.78913220399,
-      "required_pages_per_second_upper_estimate": 0.001729337261219736
+      "required_token_positions_per_second": 28252.840777363243,
+      "required_pages_per_second_upper_estimate": 0.0017204497721417735
     },
     {
       "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": 28398.78913220399,
-      "required_pages_per_second_upper_estimate": 0.0021862174002768293
+      "required_token_positions_per_second": 28252.840777363243,
+      "required_pages_per_second_upper_estimate": 0.0021749818919102847
     },
     {
       "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": 28398.78913220399,
-      "required_pages_per_second_upper_estimate": 0.35261791638879014
+      "required_token_positions_per_second": 28252.840777363243,
+      "required_pages_per_second_upper_estimate": 0.3508057262793901
     }
   ],
   "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": 0.11806987137447322,
-    "producer_p95_seconds": 0.37590185300359735,
-    "gpu_update_seconds": 0.9735626357620789,
-    "producer_batches_per_second": 8.46956118744617,
-    "gpu_batches_per_second": 1.0271552782191837,
-    "producer_headroom": 8.245648313398297,
+    "producer_batch_seconds": 0.27151926603079346,
+    "producer_p95_seconds": 0.3866263279996929,
+    "gpu_update_seconds": 0.9785918597662627,
+    "producer_batches_per_second": 3.682979902746895,
+    "gpu_batches_per_second": 1.021876474875696,
+    "producer_headroom": 3.6041341525108535,
     "sustainable": true,
-    "buffer_seconds": 60.36088341724889,
+    "buffer_seconds": 60.67269530550829,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 163,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 569212.0253505402,
+    "generation_bytes_per_second": 533541.3240771831,
     "upload_bytes_per_second": 5456928.170392238,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.06982503121926,
+    "estimated_full_shards_per_hour": 114.48554794060345,
     "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-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -336,3 +336,4 @@
 {"step": 73600, "loss": 0.6206010710448027, "lr": 0.0003, "grad_norm": 0.1478000530908829, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10771041359, "tokens_per_second": 149144.17894786637, "data_wait_s": 9.26100619835779e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 73700, "loss": 0.6197205875068903, "lr": 0.0003, "grad_norm": 0.13492088232152857, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10785605218, "tokens_per_second": 148931.69318447958, "data_wait_s": 8.645998605061322e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 73800, "loss": 0.658874835819006, "lr": 0.0003, "grad_norm": 0.15744462732086859, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10800361514, "tokens_per_second": 150753.17500461583, "data_wait_s": 1.0556999768596143e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 73900, "loss": 0.6067108102142811, "lr": 0.0003, "grad_norm": 0.12604508464039318, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10815138511, "tokens_per_second": 151301.4488370125, "data_wait_s": 1.397199957864359e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8bc0f2620e3aa5b13f1e0a5bbc9d8a9a6d3bde7ba7235f8072fff0e01b38ad6a
-size 720747
+oid sha256:81955e8d3b1d320c9c2de3c7c73df715e3e60b05f87aee805cf9846582fe35ee
+size 722865