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": 74998.87266768362,
-      "required_pages_per_second_upper_estimate": 0.006500712199929862
+      "required_token_positions_per_second": 75660.99032194329,
+      "required_pages_per_second_upper_estimate": 0.0065581028800792375
     },
     {
       "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": 9374.859083460453,
-      "required_pages_per_second_upper_estimate": 0.003739771455026509
+      "required_token_positions_per_second": 9457.623790242911,
+      "required_pages_per_second_upper_estimate": 0.0037727875340046717
     },
     {
       "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": 18749.718166920906,
-      "required_pages_per_second_upper_estimate": 0.004337417314147759
+      "required_token_positions_per_second": 18915.247580485822,
+      "required_pages_per_second_upper_estimate": 0.004375709630757822
     },
     {
       "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": 28124.57725038136,
-      "required_pages_per_second_upper_estimate": 0.0017126391962952949
+      "required_token_positions_per_second": 28372.871370728735,
+      "required_pages_per_second_upper_estimate": 0.0017277590055258783
     },
     {
       "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": 28124.57725038136,
-      "required_pages_per_second_upper_estimate": 0.002165107810547052
+      "required_token_positions_per_second": 28372.871370728735,
+      "required_pages_per_second_upper_estimate": 0.0021842221792535083
     },
     {
       "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": 28124.57725038136,
-      "required_pages_per_second_upper_estimate": 0.3492131225446858
+      "required_token_positions_per_second": 28372.871370728735,
+      "required_pages_per_second_upper_estimate": 0.3522961045324352
     }
   ],
   "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": 0.08582296080473384,
-    "producer_p95_seconds": 0.27942052999969746,
-    "gpu_update_seconds": 0.9830547763922426,
-    "producer_batches_per_second": 11.651893509887412,
-    "gpu_batches_per_second": 1.0172373137435387,
-    "producer_headroom": 11.454449568908593,
+    "producer_batch_seconds": 0.08748072325775524,
+    "producer_p95_seconds": 0.2050035009997373,
+    "gpu_update_seconds": 0.9744519558398816,
+    "producer_batches_per_second": 11.431089761953347,
+    "gpu_batches_per_second": 1.0262178591843438,
+    "producer_headroom": 11.139047775916685,
     "sustainable": true,
-    "buffer_seconds": 60.94939613631904,
+    "buffer_seconds": 60.41602126207266,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 365187.05440505676,
+    "generation_bytes_per_second": 366116.124831999,
     "upload_bytes_per_second": 6041165.722560944,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.36064075578477,
+    "estimated_full_shards_per_hour": 78.55999764175394,
     "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-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -126,3 +126,4 @@
 {"step": 127600, "loss": 0.6087887566536665, "lr": 0.0003, "grad_norm": 0.1620604392941756, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18674041976, "tokens_per_second": 149069.86359597286, "data_wait_s": 9.386001693201251e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 127700, "loss": 0.6323881652206182, "lr": 0.0003, "grad_norm": 0.14961452917126697, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18688749172, "tokens_per_second": 149980.37155267305, "data_wait_s": 1.552499998069834e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 127800, "loss": 0.5831902325153351, "lr": 0.0003, "grad_norm": 0.13881338642355742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18703401033, "tokens_per_second": 150540.16901954522, "data_wait_s": 1.344199881714303e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 127900, "loss": 0.6205321792513132, "lr": 0.0003, "grad_norm": 0.14496050650069636, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18717891551, "tokens_per_second": 147612.9805286436, "data_wait_s": 1.391300065733958e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ffe79a48f9bc1d9001728fe04a4f96af1e3be1f51e6179be1981eec50d597b4a
-size 272835
+oid sha256:6f0e20e9962af791e689a6dafb417e0a5a4759b7ba39e65a941a48484f06f4f6
+size 274953