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": 75872.28970469005,
-      "required_pages_per_second_upper_estimate": 0.006576417775042337
+      "required_token_positions_per_second": 75676.27091991807,
+      "required_pages_per_second_upper_estimate": 0.0065594273635833705
     },
     {
       "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": 9484.036213086256,
-      "required_pages_per_second_upper_estimate": 0.0037833238443777946
+      "required_token_positions_per_second": 9459.533864989759,
+      "required_pages_per_second_upper_estimate": 0.0037735494913793514
     },
     {
       "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": 18968.07242617251,
-      "required_pages_per_second_upper_estimate": 0.0043879297291219066
+      "required_token_positions_per_second": 18919.067729979517,
+      "required_pages_per_second_upper_estimate": 0.004376593355110854
     },
     {
       "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": 28452.108639258768,
-      "required_pages_per_second_upper_estimate": 0.0017325841394535382
+      "required_token_positions_per_second": 28378.601594969277,
+      "required_pages_per_second_upper_estimate": 0.0017281079461179641
     },
     {
       "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": 28452.108639258768,
-      "required_pages_per_second_upper_estimate": 0.0021903220835277483
+      "required_token_positions_per_second": 28378.601594969277,
+      "required_pages_per_second_upper_estimate": 0.0021846633077778213
     },
     {
       "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": 28452.108639258768,
-      "required_pages_per_second_upper_estimate": 0.35327996621749963
+      "required_token_positions_per_second": 28378.601594969277,
+      "required_pages_per_second_upper_estimate": 0.3523672547396759
     }
   ],
   "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": 0.07493412946868716,
-    "producer_p95_seconds": 0.2726098240018473,
-    "gpu_update_seconds": 0.9717381706412704,
-    "producer_batches_per_second": 13.34505394391579,
-    "gpu_batches_per_second": 1.0290837904824497,
-    "producer_headroom": 12.9678983065698,
+    "producer_batch_seconds": 0.3373416699608356,
+    "producer_p95_seconds": 0.3372497860000294,
+    "gpu_update_seconds": 0.9742551939169972,
+    "producer_batches_per_second": 2.9643536184429786,
+    "gpu_batches_per_second": 1.0264251155587847,
+    "producer_headroom": 2.8880369093747165,
     "sustainable": true,
-    "buffer_seconds": 60.24776657975877,
+    "buffer_seconds": 60.40382202285383,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 517929.33889289916,
-    "upload_bytes_per_second": 4008350.9631803427,
+    "generation_bytes_per_second": 476748.6250250998,
+    "upload_bytes_per_second": 3959372.8376491196,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 111.13557934847098,
+    "estimated_full_shards_per_hour": 102.29915679039712,
     "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": 0.949438960000407,
-      "bytes": 6133978,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.1755309580003086,
       "bytes": 7041796,
@@ -562,6 +556,12 @@
       "bytes": 5895199,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.0030821439977444,
+      "bytes": 1174580,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -214,3 +214,4 @@
 {"step": 136400, "loss": 0.6201961506158113, "lr": 0.0003, "grad_norm": 0.13811877783067825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19962043205, "tokens_per_second": 149190.91475862943, "data_wait_s": 1.1001000530086458e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 136500, "loss": 0.548358827829361, "lr": 0.0003, "grad_norm": 0.15211027111072187, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19976590796, "tokens_per_second": 148067.31253867495, "data_wait_s": 1.1274998541921377e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 136600, "loss": 0.617629662156105, "lr": 0.0003, "grad_norm": 0.15347404341235923, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19991072917, "tokens_per_second": 147470.402349882, "data_wait_s": 1.2537999282358214e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 136700, "loss": 0.6210627965629101, "lr": 0.0003, "grad_norm": 0.15209265436022484, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20005535906, "tokens_per_second": 147356.34552390876, "data_wait_s": 1.413200152455829e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cb3c7757bc9bc33ad5b464af7ef632930c222981a86c885737cf7179a835c665
-size 460785
+oid sha256:57aa06ce6739f1d97def89320a3a559e9937adbf7f3515466c29f3a30349a12b
+size 462903