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": 75139.30413486045,
-      "required_pages_per_second_upper_estimate": 0.00651288444358445
+      "required_token_positions_per_second": 75328.59792408576,
+      "required_pages_per_second_upper_estimate": 0.006529291949473777
     },
     {
       "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": 9392.413016857556,
-      "required_pages_per_second_upper_estimate": 0.003746773981513306
+      "required_token_positions_per_second": 9416.07474051072,
+      "required_pages_per_second_upper_estimate": 0.003756212996852848
     },
     {
       "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": 18784.82603371511,
-      "required_pages_per_second_upper_estimate": 0.004345538901253246
+      "required_token_positions_per_second": 18832.14948102144,
+      "required_pages_per_second_upper_estimate": 0.004356486347923338
     },
     {
       "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": 28177.239050572665,
-      "required_pages_per_second_upper_estimate": 0.0017158460236318329
+      "required_token_positions_per_second": 28248.22422153216,
+      "required_pages_per_second_upper_estimate": 0.001720168648112857
     },
     {
       "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": 28177.239050572665,
-      "required_pages_per_second_upper_estimate": 0.0021691618617029717
+      "required_token_positions_per_second": 28248.22422153216,
+      "required_pages_per_second_upper_estimate": 0.002174626496662963
     },
     {
       "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": 28177.239050572665,
-      "required_pages_per_second_upper_estimate": 0.34986700585535424
+      "required_token_positions_per_second": 28248.22422153216,
+      "required_pages_per_second_upper_estimate": 0.35074840410658653
     }
   ],
   "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": 0.07378440865623048,
-    "producer_p95_seconds": 0.19040167699949961,
-    "gpu_update_seconds": 0.9812174979378645,
-    "producer_batches_per_second": 13.552998773211124,
-    "gpu_batches_per_second": 1.0191420374194395,
-    "producer_headroom": 13.298439545805167,
+    "producer_batch_seconds": 0.05206777698437293,
+    "producer_p95_seconds": 0.18556611599979078,
+    "gpu_update_seconds": 0.9787517892514235,
+    "producer_batches_per_second": 19.20573640584136,
+    "gpu_batches_per_second": 1.0217094987533333,
+    "producer_headroom": 18.797648871108436,
     "sustainable": true,
-    "buffer_seconds": 60.835484872147596,
+    "buffer_seconds": 60.68261093358826,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 508888.9365283306,
+    "generation_bytes_per_second": 502315.4156123169,
     "upload_bytes_per_second": 5617612.59309987,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 109.19571945083082,
+    "estimated_full_shards_per_hour": 107.78519488598948,
     "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

@@ -65,3 +65,4 @@
 {"step": 361500, "loss": 0.5872211307287216, "lr": 0.0003, "grad_norm": 0.17403485592445106, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52913454302, "tokens_per_second": 149950.54574341545, "data_wait_s": 1.1433000508986879e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 361600, "loss": 0.6104452330619097, "lr": 0.0003, "grad_norm": 0.18630582541533516, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52928217372, "tokens_per_second": 150174.26182302722, "data_wait_s": 1.39579997266992e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 361700, "loss": 0.6090059615671635, "lr": 0.0003, "grad_norm": 0.16071528306354366, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52943041541, "tokens_per_second": 151162.04550983655, "data_wait_s": 1.5300000086426735e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 361800, "loss": 0.5769853852689266, "lr": 0.0003, "grad_norm": 0.17652536240169736, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52957703659, "tokens_per_second": 149332.9835808498, "data_wait_s": 1.167800019175047e-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 sha256:3e217cbffeec938113547c117f7370b53d1515f7761f3c79033bad5b6e1bf67f
-size 142854
+oid sha256:e23213c2c2abc50bb6caacba21ab84c4903cd0f7da96543f46339b6bd7e6ad37
+size 144972