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": 75231.43589751417,
-      "required_pages_per_second_upper_estimate": 0.005204685296371122
+      "required_token_positions_per_second": 75553.0115679572,
+      "required_pages_per_second_upper_estimate": 0.0052269326474107276
     },
     {
       "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": 9403.92948718927,
-      "required_pages_per_second_upper_estimate": 0.0039372288999077526
+      "required_token_positions_per_second": 9444.12644599465,
+      "required_pages_per_second_upper_estimate": 0.003954058527151613
     },
     {
       "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": 18807.85897437854,
-      "required_pages_per_second_upper_estimate": 0.003917181027767569
+      "required_token_positions_per_second": 18888.2528919893,
+      "required_pages_per_second_upper_estimate": 0.003933924960675711
     },
     {
       "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": 28211.788461567812,
-      "required_pages_per_second_upper_estimate": 0.0016912536413642434
+      "required_token_positions_per_second": 28332.379337983948,
+      "required_pages_per_second_upper_estimate": 0.001698482880273784
     },
     {
       "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": 28211.788461567812,
-      "required_pages_per_second_upper_estimate": 0.002193342462343748
+      "required_token_positions_per_second": 28332.379337983948,
+      "required_pages_per_second_upper_estimate": 0.0022027178725618937
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28211.788461567812,
-      "required_pages_per_second_upper_estimate": 0.3618705309265891
+      "required_token_positions_per_second": 28332.379337983948,
+      "required_pages_per_second_upper_estimate": 0.36341734120886016
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.5632282393984624,
-    "producer_p95_seconds": 0.18776499500017962,
-    "gpu_update_seconds": 0.9800158553458655,
-    "producer_batches_per_second": 1.7754791575578976,
-    "gpu_batches_per_second": 1.020391654425919,
-    "producer_headroom": 1.7399977252428598,
+    "producer_batch_seconds": 0.10519756850779771,
+    "producer_p95_seconds": 0.18183907999991789,
+    "gpu_update_seconds": 0.9758446218081503,
+    "producer_batches_per_second": 9.505923132870468,
+    "gpu_batches_per_second": 1.02475330360185,
+    "producer_headroom": 9.276303964533328,
     "sustainable": true,
-    "buffer_seconds": 60.760983031443665,
+    "buffer_seconds": 60.50236655210532,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 358352.0221448881,
-    "upload_bytes_per_second": 3809992.7171074897,
+    "generation_bytes_per_second": 357881.3978013947,
+    "upload_bytes_per_second": 3955983.9255709085,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.8940019441603,
+    "estimated_full_shards_per_hour": 76.79301691562061,
     "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": [
@@ -280,6 +280,12 @@
       "bytes": 6120972,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.693875957000273,
+      "bytes": 10707952,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -60,3 +60,4 @@
 {"step": 806000, "loss": 0.5180491209030151, "lr": 3e-05, "grad_norm": 0.25843856561638007, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117974916154, "tokens_per_second": 151817.38622671983, "data_wait_s": 1.479599995946046e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 806100, "loss": 0.5533642172813416, "lr": 3e-05, "grad_norm": 0.2787937548759742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117989609136, "tokens_per_second": 151631.99965050165, "data_wait_s": 1.6586999663559254e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 806200, "loss": 0.5670785903930664, "lr": 3e-05, "grad_norm": 0.3036575457163, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118004391654, "tokens_per_second": 152700.3048451993, "data_wait_s": 1.4807999832555652e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 806300, "loss": 0.5046426653862, "lr": 3e-05, "grad_norm": 0.27817865482446136, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118018968740, "tokens_per_second": 150404.6111798633, "data_wait_s": 1.2620000234164763e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8fa0b3c8bd2805bc05316cc9c977e7b092bbe9853258259d9b811c522d833745
-size 132264
+oid sha256:6851cdb3ad507aef686dd78573747345d0bf0188be4707dde4b7ae6908809a2b
+size 134382