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": 75329.11845854671,
-      "required_pages_per_second_upper_estimate": 0.00652933706808157
+      "required_token_positions_per_second": 76065.0880979941,
+      "required_pages_per_second_upper_estimate": 0.006593129051130872
     },
     {
       "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": 9416.139807318339,
-      "required_pages_per_second_upper_estimate": 0.003756238952975243
+      "required_token_positions_per_second": 9508.136012249262,
+      "required_pages_per_second_upper_estimate": 0.0037929376145880256
     },
     {
       "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": 18832.279614636678,
-      "required_pages_per_second_upper_estimate": 0.0043565164520449505
+      "required_token_positions_per_second": 19016.272024498525,
+      "required_pages_per_second_upper_estimate": 0.0043990798579106715
     },
     {
       "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": 28248.41942195502,
-      "required_pages_per_second_upper_estimate": 0.0017201805347944717
+      "required_token_positions_per_second": 28524.408036747787,
+      "required_pages_per_second_upper_estimate": 0.0017369867934349968
     },
     {
       "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": 28248.41942195502,
-      "required_pages_per_second_upper_estimate": 0.0021746415237318645
+      "required_token_positions_per_second": 28524.408036747787,
+      "required_pages_per_second_upper_estimate": 0.0021958878912839946
     },
     {
       "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": 28248.41942195502,
-      "required_pages_per_second_upper_estimate": 0.35075082784254463
+      "required_token_positions_per_second": 28524.408036747787,
+      "required_pages_per_second_upper_estimate": 0.35417768276379535
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.09673103116418247,
-    "producer_p95_seconds": 0.21955628000068828,
-    "gpu_update_seconds": 0.9787450259433501,
-    "producer_batches_per_second": 10.337944173289033,
-    "gpu_batches_per_second": 1.0217165589538129,
-    "producer_headroom": 10.11821143808668,
+    "producer_batch_seconds": 0.08934853077349203,
+    "producer_p95_seconds": 0.21759994499734603,
+    "gpu_update_seconds": 0.969275154260214,
+    "producer_batches_per_second": 11.192125839596688,
+    "gpu_batches_per_second": 1.0316987860513522,
+    "producer_headroom": 10.848249499674807,
     "sustainable": true,
-    "buffer_seconds": 60.682191608487706,
+    "buffer_seconds": 60.09505956413327,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 326111.9211780653,
-    "upload_bytes_per_second": 5160657.26484681,
+    "generation_bytes_per_second": 346725.0642022869,
+    "upload_bytes_per_second": 5182151.64904447,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.97602678781956,
+    "estimated_full_shards_per_hour": 74.39912743140654,
     "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": 1.4731977419996838,
-      "bytes": 6241189,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.5922990050003136,
       "bytes": 11693360,
@@ -562,6 +556,12 @@
       "bytes": 8515706,
       "pages": 41,
       "shards": 4
+    },
+    {
+      "seconds": 1.1056661020011234,
+      "bytes": 6106174,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -169,3 +169,4 @@
 {"step": 171900, "loss": 0.6244958490133286, "lr": 0.0003, "grad_norm": 0.14701528131234995, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25158621255, "tokens_per_second": 148934.09620441706, "data_wait_s": 1.2003001756966114e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 172000, "loss": 0.6820664405822754, "lr": 0.0003, "grad_norm": 0.14046266448044453, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25173218574, "tokens_per_second": 149847.75509250868, "data_wait_s": 1.0988002031808719e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 172100, "loss": 0.5721213407814503, "lr": 0.0003, "grad_norm": 0.14904676025056426, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25187867206, "tokens_per_second": 149166.60639444797, "data_wait_s": 9.974999557016417e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 172200, "loss": 0.6270482540130615, "lr": 0.0003, "grad_norm": 0.13754360075202146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25202419943, "tokens_per_second": 140488.59696334766, "data_wait_s": 1.0435996955493465e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9d3327a0efb2e5cb0c0daa57fecb64c95b40554fdbb14ceca215b63a4ec24613
-size 364692
+oid sha256:1c97ce4f4e2d65b6a9e313283d3809c9cafecf153c3379c51472c9d957eb71e7
+size 366810