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": 77429.38397843782,
-      "required_pages_per_second_upper_estimate": 0.005356744444551603
+      "required_token_positions_per_second": 77638.75610037129,
+      "required_pages_per_second_upper_estimate": 0.005371229293757215
     },
     {
       "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": 9678.672997304728,
-      "required_pages_per_second_upper_estimate": 0.004052258270296194
+      "required_token_positions_per_second": 9704.844512546411,
+      "required_pages_per_second_upper_estimate": 0.004063215737204501
     },
     {
       "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": 19357.345994609455,
-      "required_pages_per_second_upper_estimate": 0.004031624683134471
+      "required_token_positions_per_second": 19409.689025092823,
+      "required_pages_per_second_upper_estimate": 0.004042526356005617
     },
     {
       "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": 29036.01899191418,
-      "required_pages_per_second_upper_estimate": 0.0017406650031313595
+      "required_token_positions_per_second": 29114.533537639236,
+      "required_pages_per_second_upper_estimate": 0.0017453718302628068
     },
     {
       "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": 29036.01899191418,
-      "required_pages_per_second_upper_estimate": 0.00225742276067122
+      "required_token_positions_per_second": 29114.533537639236,
+      "required_pages_per_second_upper_estimate": 0.002263526921252354
     },
     {
       "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": 29036.01899191418,
-      "required_pages_per_second_upper_estimate": 0.37244287518007957
+      "required_token_positions_per_second": 29114.533537639236,
+      "required_pages_per_second_upper_estimate": 0.37344997547028946
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1190755792656546,
-    "producer_p95_seconds": 0.31447203300194815,
-    "gpu_update_seconds": 0.9521966495372278,
-    "producer_batches_per_second": 8.398027590267064,
-    "gpu_batches_per_second": 1.0502032332144886,
-    "producer_headroom": 7.996573734173497,
+    "producer_batch_seconds": 0.44102988181197134,
+    "producer_p95_seconds": 0.3099656300037168,
+    "gpu_update_seconds": 0.9496288155967431,
+    "producer_batches_per_second": 2.267420057551429,
+    "gpu_batches_per_second": 1.053043024364845,
+    "producer_headroom": 2.1532074237128627,
     "sustainable": true,
-    "buffer_seconds": 60.94058557038258,
+    "buffer_seconds": 60.776244198191556,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 345438.348312401,
-    "upload_bytes_per_second": 5036813.362771684,
+    "generation_bytes_per_second": 310645.1198192576,
+    "upload_bytes_per_second": 5068666.151359088,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.12302815464994,
+    "estimated_full_shards_per_hour": 66.65721126492782,
     "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": 1.9872482999999193,
-      "bytes": 7546218,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.5842436470011307,
       "bytes": 5584785,
@@ -562,6 +556,12 @@
       "bytes": 7683555,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.031134179997025,
+      "bytes": 5438180,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -380,3 +380,4 @@
 {"step": 923000, "loss": 0.49429020285606384, "lr": 3e-05, "grad_norm": 0.32696403912165645, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135100832265, "tokens_per_second": 152626.65675231628, "data_wait_s": 1.3780998415313661e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 923100, "loss": 0.5448967218399048, "lr": 3e-05, "grad_norm": 0.3292880024477215, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135115520222, "tokens_per_second": 153429.7172407346, "data_wait_s": 8.725000952836126e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 923200, "loss": 0.5117206573486328, "lr": 3e-05, "grad_norm": 0.3306566202676778, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135130219494, "tokens_per_second": 153538.5503625293, "data_wait_s": 1.2519994925241917e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 923300, "loss": 0.5540512800216675, "lr": 3e-05, "grad_norm": 0.33279744847035647, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135144911724, "tokens_per_second": 152360.5292352392, "data_wait_s": 1.515400072094053e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:424df42d64a929ae4c0d7405b1d228a506cb072c26d4d65fdfe61effb72c882d
-size 814722
+oid sha256:919a174c605d20fd4beb04e8e97e0f87bc5bd7046d544fd4cf060259f03805f0
+size 816840