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": 75476.28258636278,
-      "required_pages_per_second_upper_estimate": 0.006542092881696595
+      "required_token_positions_per_second": 75826.94898411506,
+      "required_pages_per_second_upper_estimate": 0.006572487756297904
     },
     {
       "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": 9434.535323295348,
-      "required_pages_per_second_upper_estimate": 0.003763577199335945
+      "required_token_positions_per_second": 9478.368623014383,
+      "required_pages_per_second_upper_estimate": 0.003781062957960102
     },
     {
       "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": 18869.070646590695,
-      "required_pages_per_second_upper_estimate": 0.004365027409787465
+      "required_token_positions_per_second": 18956.737246028766,
+      "required_pages_per_second_upper_estimate": 0.004385307534688008
     },
     {
       "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": 28303.605969886044,
-      "required_pages_per_second_upper_estimate": 0.001723541106022031
+      "required_token_positions_per_second": 28435.105869043153,
+      "required_pages_per_second_upper_estimate": 0.0017315487599540483
     },
     {
       "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": 28303.605969886044,
-      "required_pages_per_second_upper_estimate": 0.002178889936957737
+      "required_token_positions_per_second": 28435.105869043153,
+      "required_pages_per_second_upper_estimate": 0.0021890131632098697
     },
     {
       "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": 28303.605969886044,
-      "required_pages_per_second_upper_estimate": 0.35143606007035333
+      "required_token_positions_per_second": 28435.105869043153,
+      "required_pages_per_second_upper_estimate": 0.35306884871603306
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.4783490578359988,
-    "producer_p95_seconds": 0.3903906810010085,
-    "gpu_update_seconds": 0.9768366627706878,
-    "producer_batches_per_second": 2.0905236116151156,
-    "gpu_batches_per_second": 1.023712600183957,
-    "producer_headroom": 2.042100108213435,
+    "producer_batch_seconds": 0.12061077685910959,
+    "producer_p95_seconds": 0.37974605199997313,
+    "gpu_update_seconds": 0.9723192214346541,
+    "producer_batches_per_second": 8.291133064901333,
+    "gpu_batches_per_second": 1.0284688176013872,
+    "producer_headroom": 8.061628046475981,
     "sustainable": true,
-    "buffer_seconds": 60.563873091782646,
+    "buffer_seconds": 60.28379172894856,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 74,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 263092.39071651525,
-    "upload_bytes_per_second": 5303738.313631662,
+    "generation_bytes_per_second": 252078.8854392971,
+    "upload_bytes_per_second": 5376933.8780853655,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.45350257035821,
+    "estimated_full_shards_per_hour": 54.09026071914849,
     "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": 1.2800720170016575,
-      "bytes": 6017394,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0338583490010933,
       "bytes": 6530808,
@@ -562,6 +556,12 @@
       "bytes": 5030935,
       "pages": 25,
       "shards": 3
+    },
+    {
+      "seconds": 1.2526760300024762,
+      "bytes": 11691054,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -345,3 +345,4 @@
 {"step": 389500, "loss": 0.5847944058477879, "lr": 0.0003, "grad_norm": 0.1547880307632339, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57012298598, "tokens_per_second": 138880.10867600585, "data_wait_s": 1.112099562305957e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 389600, "loss": 0.5617741234600544, "lr": 0.0003, "grad_norm": 0.14156084775089878, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57026864321, "tokens_per_second": 128063.9472956184, "data_wait_s": 1.4544995792675763e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 389700, "loss": 0.6059103906154633, "lr": 0.0003, "grad_norm": 0.15986029536768465, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57041425564, "tokens_per_second": 148058.45425343263, "data_wait_s": 1.1693999113049358e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 389800, "loss": 0.6095029674470425, "lr": 0.0003, "grad_norm": 0.17847424212721572, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57056045820, "tokens_per_second": 148399.3938685122, "data_wait_s": 9.235998732037842e-06, "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 sha256:0d18f988baec1ac41c7d663355ead4e0fe57440dad792e65c0e421d151014b0d
-size 739809
+oid sha256:019491f6705aa4d620fce35df4674dbaee2cffb60f980ec93c4ed0c4f20633c6
+size 741927