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": 76528.58935689517,
-      "required_pages_per_second_upper_estimate": 0.005294425382501797
+      "required_token_positions_per_second": 77300.98167934327,
+      "required_pages_per_second_upper_estimate": 0.005347861275565861
     },
     {
       "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": 9566.073669611897,
-      "required_pages_per_second_upper_estimate": 0.004005115283132547
+      "required_token_positions_per_second": 9662.622709917909,
+      "required_pages_per_second_upper_estimate": 0.004045538350135446
     },
     {
       "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": 19132.147339223793,
-      "required_pages_per_second_upper_estimate": 0.003984721742105553
+      "required_token_positions_per_second": 19325.245419835817,
+      "required_pages_per_second_upper_estimate": 0.004024938979958207
     },
     {
       "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": 28698.221008835688,
-      "required_pages_per_second_upper_estimate": 0.0017204145298334588
+      "required_token_positions_per_second": 28987.868129753726,
+      "required_pages_per_second_upper_estimate": 0.0017377784324669496
     },
     {
       "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": 28698.221008835688,
-      "required_pages_per_second_upper_estimate": 0.0022311604533031687
+      "required_token_positions_per_second": 28987.868129753726,
+      "required_pages_per_second_upper_estimate": 0.002253679242931504
     },
     {
       "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": 28698.221008835688,
-      "required_pages_per_second_upper_estimate": 0.3681099653523645
+      "required_token_positions_per_second": 28987.868129753726,
+      "required_pages_per_second_upper_estimate": 0.37182524762065294
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.5082066507266063,
-    "producer_p95_seconds": 0.17077684299965767,
-    "gpu_update_seconds": 0.9634046651005879,
-    "producer_batches_per_second": 1.9677034894570038,
-    "gpu_batches_per_second": 1.0379854242200408,
-    "producer_headroom": 1.895694721277583,
+    "producer_batch_seconds": 0.2756291726641322,
+    "producer_p95_seconds": 0.16944977099956304,
+    "gpu_update_seconds": 0.9537783143018214,
+    "producer_batches_per_second": 3.628062988885975,
+    "gpu_batches_per_second": 1.0484616655726897,
+    "producer_headroom": 3.460367801720493,
     "sustainable": true,
-    "buffer_seconds": 60.69449390133703,
+    "buffer_seconds": 60.088033801014745,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 312314.1376016931,
-    "upload_bytes_per_second": 5741509.716849434,
+    "generation_bytes_per_second": 297692.3094708694,
+    "upload_bytes_per_second": 5922209.00531333,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.01534362829298,
+    "estimated_full_shards_per_hour": 63.87783969015658,
     "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": [
@@ -298,6 +298,12 @@
       "bytes": 2583429,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.304920979000599,
+      "bytes": 11842449,
+      "pages": 27,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -28,3 +28,4 @@
 {"step": 487800, "loss": 0.5893129110336304, "lr": 0.0003, "grad_norm": 0.15771968225877409, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71397697664, "tokens_per_second": 152379.4101436609, "data_wait_s": 1.1095999980170745e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 487900, "loss": 0.5991864204406738, "lr": 0.0003, "grad_norm": 0.1773661234737318, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71412514648, "tokens_per_second": 136685.44896800318, "data_wait_s": 1.4724000720889308e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 488000, "loss": 0.5991197228431702, "lr": 0.0003, "grad_norm": 0.15802435942714238, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71427219379, "tokens_per_second": 153058.96275634225, "data_wait_s": 1.2954999874636997e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 488100, "loss": 0.6014329195022583, "lr": 0.0003, "grad_norm": 0.20564692473549886, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71441950975, "tokens_per_second": 154479.18494264406, "data_wait_s": 9.575000149197876e-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:e7d396365164283403b7dbb9f1f293a79c4ba1cf4a86a02a173a97d655e2b227
-size 63705
+oid sha256:39493cf93f69ecf508bd7de6b03ab3206fd6bb49ba021b4df5624694439f4d9a
+size 65823