Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76653.34070687914,
-      "required_pages_per_second_upper_estimate": 0.0053030559703567005
+      "required_token_positions_per_second": 77072.03269090403,
+      "required_pages_per_second_upper_estimate": 0.005332022053310791
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9581.667588359893,
-      "required_pages_per_second_upper_estimate": 0.0040116441312742805
+      "required_token_positions_per_second": 9634.004086363004,
+      "required_pages_per_second_upper_estimate": 0.004033556330077826
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19163.335176719786,
-      "required_pages_per_second_upper_estimate": 0.0039912173461773845
+      "required_token_positions_per_second": 19268.008172726008,
+      "required_pages_per_second_upper_estimate": 0.004013017970832946
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28745.00276507968,
-      "required_pages_per_second_upper_estimate": 0.0017232190246886804
+      "required_token_positions_per_second": 28902.012259089013,
+      "required_pages_per_second_upper_estimate": 0.001732631504114924
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28745.00276507968,
-      "required_pages_per_second_upper_estimate": 0.0022347975290799398
+      "required_token_positions_per_second": 28902.012259089013,
+      "required_pages_per_second_upper_estimate": 0.0022470043266273874
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28745.00276507968,
-      "required_pages_per_second_upper_estimate": 0.3687100314911261
+      "required_token_positions_per_second": 28902.012259089013,
+      "required_pages_per_second_upper_estimate": 0.3707239806966177
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.5414521049997347,
-    "producer_p95_seconds": 0.21426030400107265,
-    "gpu_update_seconds": 0.9618367486674118,
-    "producer_batches_per_second": 1.8468854230430778,
-    "gpu_batches_per_second": 1.0396774726953009,
-    "producer_headroom": 1.7764022704609912,
+    "producer_batch_seconds": 0.0816099174686542,
+    "producer_p95_seconds": 0.21490330300002825,
+    "gpu_update_seconds": 0.9566115934126816,
+    "producer_batches_per_second": 12.253412710337479,
+    "gpu_batches_per_second": 1.0453563461765414,
+    "producer_headroom": 11.72175665757914,
     "sustainable": true,
-    "buffer_seconds": 60.59571516604694,
+    "buffer_seconds": 60.266530384998944,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 329843.48533924384,
-    "upload_bytes_per_second": 4220376.1224243,
+    "generation_bytes_per_second": 341746.6792897314,
+    "upload_bytes_per_second": 4283830.6711375825,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.77673359044061,
+    "estimated_full_shards_per_hour": 73.33088192004162,
     "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": 1.1851188379987434,
-      "bytes": 1113462,
-      "pages": 32,
-      "shards": 1
-    },
     {
       "seconds": 1.064678984999773,
       "bytes": 6185154,
@@ -562,6 +556,12 @@
       "bytes": 9910776,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 0.9998362250007631,
+      "bytes": 6697585,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -167,3 +167,4 @@
 {"step": 701700, "loss": 0.5553956031799316, "lr": 8e-05, "grad_norm": 0.20827242176660823, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102709466655, "tokens_per_second": 152349.58702969798, "data_wait_s": 1.572099790791981e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 701800, "loss": 0.4930936098098755, "lr": 8e-05, "grad_norm": 0.22013987717921163, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102723854050, "tokens_per_second": 115924.32924219397, "data_wait_s": 1.7975999071495607e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 701900, "loss": 0.5594791173934937, "lr": 8e-05, "grad_norm": 0.21986700001083642, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102738419199, "tokens_per_second": 152021.83071301106, "data_wait_s": 1.4746001397725195e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 702000, "loss": 0.5378317832946777, "lr": 8e-05, "grad_norm": 0.21696843459342713, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102753059104, "tokens_per_second": 152453.74243015185, "data_wait_s": 1.5072000678628683e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:709a7bbf7a8b14868b422ce55c5a6ac5ad87d30009a98587bb6df2594e6b14f4
-size 360456
+oid sha256:2e1c5844052f4b4e1da9fc5b7b4eca9d1ef1af7006990fe3e9e09fdd27ef13c1
+size 362574