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": 74811.10529717195,
-      "required_pages_per_second_upper_estimate": 0.0064844370001459605
+      "required_token_positions_per_second": 74385.04416725476,
+      "required_pages_per_second_upper_estimate": 0.0064475071012468
     },
     {
       "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": 9351.388162146493,
-      "required_pages_per_second_upper_estimate": 0.003730408553592825
+      "required_token_positions_per_second": 9298.130520906845,
+      "required_pages_per_second_upper_estimate": 0.0037091632842296334
     },
     {
       "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": 18702.776324292987,
-      "required_pages_per_second_upper_estimate": 0.004326558145004004
+      "required_token_positions_per_second": 18596.26104181369,
+      "required_pages_per_second_upper_estimate": 0.004301917709007365
     },
     {
       "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": 28054.16448643948,
-      "required_pages_per_second_upper_estimate": 0.001708351428398456
+      "required_token_positions_per_second": 27894.391562720535,
+      "required_pages_per_second_upper_estimate": 0.0016986220956077185
     },
     {
       "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": 28054.16448643948,
-      "required_pages_per_second_upper_estimate": 0.0021596872410638004
+      "required_token_positions_per_second": 27894.391562720535,
+      "required_pages_per_second_upper_estimate": 0.0021473874791161558
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28054.16448643948,
-      "required_pages_per_second_upper_estimate": 0.34833883167288926
+      "required_token_positions_per_second": 27894.391562720535,
+      "required_pages_per_second_upper_estimate": 0.3463549866858777
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11486566470273374,
-    "producer_p95_seconds": 0.3623625760010327,
-    "gpu_update_seconds": 0.9855221321370734,
-    "producer_batches_per_second": 8.70582173174157,
-    "gpu_batches_per_second": 1.014690555788465,
-    "producer_headroom": 8.57977999507122,
+    "producer_batch_seconds": 0.12566787948429692,
+    "producer_p95_seconds": 0.3616767479979899,
+    "gpu_update_seconds": 0.991166985586815,
+    "producer_batches_per_second": 7.957482883483818,
+    "gpu_batches_per_second": 1.0089117318692322,
+    "producer_headroom": 7.887194322481333,
     "sustainable": true,
-    "buffer_seconds": 60.11685006036148,
+    "buffer_seconds": 60.461186120795716,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 416725.0325285504,
-    "upload_bytes_per_second": 6010721.471126111,
+    "generation_bytes_per_second": 458147.7054689269,
+    "upload_bytes_per_second": 5997698.200850242,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.41949111835847,
+    "estimated_full_shards_per_hour": 98.30783246088845,
     "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": 0.995095141002821,
-      "bytes": 6689673,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 0.9397628990009252,
       "bytes": 7665294,
@@ -562,6 +556,12 @@
       "bytes": 915809,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.023697215001448,
+      "bytes": 5944681,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -300,3 +300,4 @@
 {"step": 115000, "loss": 0.6215994767844677, "lr": 0.0003, "grad_norm": 0.14603409049989743, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16828383176, "tokens_per_second": 146871.65805112463, "data_wait_s": 1.9731000065803528e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 115100, "loss": 0.6090226266533136, "lr": 0.0003, "grad_norm": 0.12995592805320222, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16843057721, "tokens_per_second": 133816.77263533513, "data_wait_s": 1.7527003365103155e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 115200, "loss": 0.5843383781611919, "lr": 0.0003, "grad_norm": 0.13943145174520966, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16857707507, "tokens_per_second": 149062.55401111444, "data_wait_s": 1.1841999366879463e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 115300, "loss": 0.6172855384647846, "lr": 0.0003, "grad_norm": 0.12460532216136234, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16872235692, "tokens_per_second": 148134.06387533378, "data_wait_s": 1.2730000889860094e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:087aa4cd388e496fdd7c2c312bf2f6daacb14f84217414db3a48e6e322e7a0ec
-size 644499
+oid sha256:c6b258d98ae18acf24f6ff2f64e7b51d83e3a18bd0ab118f50cb8671d8d7c982
+size 646617