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": 75476.10452085512,
-      "required_pages_per_second_upper_estimate": 0.006542077447429694
+      "required_token_positions_per_second": 74520.23167839296,
+      "required_pages_per_second_upper_estimate": 0.006459224812082645
     },
     {
       "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": 9434.51306510689,
-      "required_pages_per_second_upper_estimate": 0.00376356832021178
+      "required_token_positions_per_second": 9315.02895979912,
+      "required_pages_per_second_upper_estimate": 0.0037159043241579382
     },
     {
       "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": 18869.02613021378,
-      "required_pages_per_second_upper_estimate": 0.004365017111707127
+      "required_token_positions_per_second": 18630.05791959824,
+      "required_pages_per_second_upper_estimate": 0.004309736021878086
     },
     {
       "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": 28303.53919532067,
-      "required_pages_per_second_upper_estimate": 0.001723537039801339
+      "required_token_positions_per_second": 27945.08687939736,
+      "required_pages_per_second_upper_estimate": 0.0017017091744155662
     },
     {
       "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": 28303.53919532067,
-      "required_pages_per_second_upper_estimate": 0.0021788847964668507
+      "required_token_positions_per_second": 27945.08687939736,
+      "required_pages_per_second_upper_estimate": 0.002151290144927556
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28303.53919532067,
-      "required_pages_per_second_upper_estimate": 0.35143523095373147
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27945.08687939736,
+      "required_pages_per_second_upper_estimate": 0.3469844528526933
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.07874541979705896,
-    "producer_p95_seconds": 0.27901098299844307,
-    "gpu_update_seconds": 0.9768389673532755,
-    "producer_batches_per_second": 12.699151297652346,
-    "gpu_batches_per_second": 1.0237101850159385,
-    "producer_headroom": 12.405025839861725,
+    "producer_batch_seconds": 0.20415187885922137,
+    "producer_p95_seconds": 0.3456205590009631,
+    "gpu_update_seconds": 0.9893689047853207,
+    "producer_batches_per_second": 4.898313968932796,
+    "gpu_batches_per_second": 1.010745329839314,
+    "producer_headroom": 4.846239526737678,
     "sustainable": true,
-    "buffer_seconds": 60.56401597590308,
+    "buffer_seconds": 60.35150319190456,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 364662.6653252356,
+    "generation_bytes_per_second": 393190.28671041597,
     "upload_bytes_per_second": 5204026.453271514,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.24811906640817,
+    "estimated_full_shards_per_hour": 84.36948252663001,
     "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-v1/1790738827394728502-c744ae20/train.jsonl

@@ -249,3 +249,4 @@
 {"step": 339900, "loss": 0.5688153691589832, "lr": 0.0003, "grad_norm": 0.15310673850184375, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49752329398, "tokens_per_second": 150021.12086090905, "data_wait_s": 1.387000156682916e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 340000, "loss": 0.561069430783391, "lr": 0.0003, "grad_norm": 0.16010040769861256, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49766917329, "tokens_per_second": 148441.14362928597, "data_wait_s": 1.1295000149402767e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 340100, "loss": 0.589220168069005, "lr": 0.0003, "grad_norm": 0.1648770940891779, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49781441091, "tokens_per_second": 135025.60458267588, "data_wait_s": 1.3221000699559227e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 340200, "loss": 0.5384630095213652, "lr": 0.0003, "grad_norm": 0.14425019723219684, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49795998935, "tokens_per_second": 148280.82035737755, "data_wait_s": 1.2831998901674524e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:79ff44345f4eed4b75b294ae0d2b87a04d09c862c97f7d84b89c7af06b749f9a
-size 535698
+oid sha256:cfa8cd9a44055c158be2116f78e13d49b10171ad9f0d3869a2912dbefcc8f18e
+size 537816