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": 75816.59753640718,
-      "required_pages_per_second_upper_estimate": 0.006571590519046089
+      "required_token_positions_per_second": 75571.83458993434,
+      "required_pages_per_second_upper_estimate": 0.0065503750871918395
     },
     {
       "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": 9477.074692050897,
-      "required_pages_per_second_upper_estimate": 0.0037805467895527752
+      "required_token_positions_per_second": 9446.479323741793,
+      "required_pages_per_second_upper_estimate": 0.0037683418396927525
     },
     {
       "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": 18954.149384101795,
-      "required_pages_per_second_upper_estimate": 0.004384708878376018
+      "required_token_positions_per_second": 18892.958647483585,
+      "required_pages_per_second_upper_estimate": 0.004370553478379578
     },
     {
       "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": 28431.224076152695,
-      "required_pages_per_second_upper_estimate": 0.001731312379133213
+      "required_token_positions_per_second": 28339.437971225376,
+      "required_pages_per_second_upper_estimate": 0.0017257230869076143
     },
     {
       "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": 28431.224076152695,
-      "required_pages_per_second_upper_estimate": 0.002188714332047672
+      "required_token_positions_per_second": 28339.437971225376,
+      "required_pages_per_second_upper_estimate": 0.0021816483836101666
     },
     {
       "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": 28431.224076152695,
-      "required_pages_per_second_upper_estimate": 0.35302064983985865
+      "required_token_positions_per_second": 28339.437971225376,
+      "required_pages_per_second_upper_estimate": 0.35188097360499365
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.2189116486716216,
-    "producer_p95_seconds": 0.39063410100061446,
-    "gpu_update_seconds": 0.9724519748409411,
-    "producer_batches_per_second": 4.568052938562671,
-    "gpu_batches_per_second": 1.0283284171062173,
-    "producer_headroom": 4.442212101283233,
+    "producer_batch_seconds": 0.44650785668744675,
+    "producer_p95_seconds": 0.43059379299666034,
+    "gpu_update_seconds": 0.9756015637315238,
+    "producer_batches_per_second": 2.2396022489252525,
+    "gpu_batches_per_second": 1.0250086071768438,
+    "producer_headroom": 2.1849594561881136,
     "sustainable": true,
-    "buffer_seconds": 60.29202244013835,
+    "buffer_seconds": 60.48729695135448,
     "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": 527936.4549714911,
+    "generation_bytes_per_second": 483988.12176279345,
     "upload_bytes_per_second": 5716853.5664507225,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 113.28287350519705,
+    "estimated_full_shards_per_hour": 103.85258426344731,
     "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-v1/1790782192342215557-281f6532/train.jsonl

@@ -397,3 +397,4 @@
 {"step": 394700, "loss": 0.5499810837209225, "lr": 0.0003, "grad_norm": 0.1470346524536101, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57772545742, "tokens_per_second": 149039.21257245296, "data_wait_s": 9.467999916523695e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 394800, "loss": 0.593715151771903, "lr": 0.0003, "grad_norm": 0.17916264473158008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57787309955, "tokens_per_second": 151255.48103643116, "data_wait_s": 1.2794000213034451e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 394900, "loss": 0.6093955338001251, "lr": 0.0003, "grad_norm": 0.16123570419690397, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57802033460, "tokens_per_second": 149377.65332230096, "data_wait_s": 1.1211996024940163e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 395000, "loss": 0.5624208748340607, "lr": 0.0003, "grad_norm": 0.14423838606737646, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57816763361, "tokens_per_second": 150025.41638299677, "data_wait_s": 1.1039002856705338e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5c238ce268df066ef25dff4ea061a2e8b4c89b83f32c3a7d339aebeec9bcac5b
-size 850728
+oid sha256:b4cc7180adecf8130f85eb81b0dd58fc55de7005129d99410413db39f6e90664
+size 852846