Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77213.6919826638,
-      "required_pages_per_second_upper_estimate": 0.005341822371809575
+      "required_token_positions_per_second": 76718.48697520119,
+      "required_pages_per_second_upper_estimate": 0.005307562940359388
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9651.711497832976,
-      "required_pages_per_second_upper_estimate": 0.004040970053487503
+      "required_token_positions_per_second": 9589.810871900148,
+      "required_pages_per_second_upper_estimate": 0.004015053554041488
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19303.42299566595,
-      "required_pages_per_second_upper_estimate": 0.00402039394449965
+      "required_token_positions_per_second": 19179.621743800297,
+      "required_pages_per_second_upper_estimate": 0.003994609408594598
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28955.13449349893,
-      "required_pages_per_second_upper_estimate": 0.001735816100954843
+      "required_token_positions_per_second": 28769.432615700447,
+      "required_pages_per_second_upper_estimate": 0.0017246835569311733
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28955.13449349893,
-      "required_pages_per_second_upper_estimate": 0.002251134346692743
+      "required_token_positions_per_second": 28769.432615700447,
+      "required_pages_per_second_upper_estimate": 0.002236696842510148
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28955.13449349893,
-      "required_pages_per_second_upper_estimate": 0.37140537568141674
+      "required_token_positions_per_second": 28769.432615700447,
+      "required_pages_per_second_upper_estimate": 0.36902339138415935
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,39 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11904607814852852,
-    "producer_p95_seconds": 0.3377903690052335,
-    "gpu_update_seconds": 0.9548565559661825,
-    "producer_batches_per_second": 8.400108727247144,
-    "gpu_batches_per_second": 1.0472777232891683,
-    "producer_headroom": 8.02089888904068,
+    "producer_batch_seconds": 0.1210615917418636,
+    "producer_p95_seconds": 0.335499847002211,
+    "gpu_update_seconds": 0.9610199953999634,
+    "producer_batches_per_second": 8.260258151340627,
+    "gpu_batches_per_second": 1.0405610755099988,
+    "producer_headroom": 7.9382732506038804,
     "sustainable": true,
-    "buffer_seconds": 60.1559630258695,
+    "buffer_seconds": 60.54425971019769,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 384588.48339903663,
-    "upload_bytes_per_second": 3837530.023601385,
+    "generation_bytes_per_second": 465917.34385255916,
+    "upload_bytes_per_second": 3795656.338218112,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.52373577574086,
+    "estimated_full_shards_per_hour": 99.97501599009115,
     "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": 1.521548144002736,
-      "bytes": 10001036,
-      "pages": 42,
-      "shards": 3
-    },
-    {
-      "seconds": 1.037408388001495,
-      "bytes": 2352962,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.4666764929934288,
       "bytes": 11848768,
@@ -562,6 +550,18 @@
       "bytes": 12058095,
       "pages": 40,
       "shards": 4
+    },
+    {
+      "seconds": 0.957104379995144,
+      "bytes": 857684,
+      "pages": 27,
+      "shards": 1
+    },
+    {
+      "seconds": 1.3987013609948917,
+      "bytes": 5890291,
+      "pages": 19,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -398,3 +398,4 @@
 {"step": 524800, "loss": 0.5686387419700623, "lr": 0.0003, "grad_norm": 0.15876198911507197, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76812983798, "tokens_per_second": 151723.5168645376, "data_wait_s": 1.5948993677739054e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 524900, "loss": 0.5441434383392334, "lr": 0.0003, "grad_norm": 0.198230476678356, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76827643313, "tokens_per_second": 152883.9950594783, "data_wait_s": 1.6022997442632914e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 525000, "loss": 0.5860968828201294, "lr": 0.0003, "grad_norm": 0.14732875713590157, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76842395418, "tokens_per_second": 153799.0157697656, "data_wait_s": 1.505100226495415e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 525100, "loss": 0.6338948011398315, "lr": 0.0003, "grad_norm": 0.16164860273831738, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76857132722, "tokens_per_second": 138385.84955087712, "data_wait_s": 9.424002200830728e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d9ccfa04cf55e98100a4f73622a51fac8e40bfc4df82c212a4e71398a0107cf1
-size 853629
+oid sha256:bcf263d11a60d3496e628ed3dcc44d72315375f5da789340a2883250ea9b091c
+size 855747