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": 75912.4606356097,
-      "required_pages_per_second_upper_estimate": 0.00657989968952216
+      "required_token_positions_per_second": 75872.28970469005,
+      "required_pages_per_second_upper_estimate": 0.006576417775042337
     },
     {
       "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": 9489.057579451213,
-      "required_pages_per_second_upper_estimate": 0.0037853269424968938
+      "required_token_positions_per_second": 9484.036213086256,
+      "required_pages_per_second_upper_estimate": 0.0037833238443777946
     },
     {
       "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": 18978.115158902427,
-      "required_pages_per_second_upper_estimate": 0.004390252938592913
+      "required_token_positions_per_second": 18968.07242617251,
+      "required_pages_per_second_upper_estimate": 0.0043879297291219066
     },
     {
       "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": 28467.17273835364,
-      "required_pages_per_second_upper_estimate": 0.0017335014640531964
+      "required_token_positions_per_second": 28452.108639258768,
+      "required_pages_per_second_upper_estimate": 0.0017325841394535382
     },
     {
       "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": 28467.17273835364,
-      "required_pages_per_second_upper_estimate": 0.002191481759576167
+      "required_token_positions_per_second": 28452.108639258768,
+      "required_pages_per_second_upper_estimate": 0.0021903220835277483
     },
     {
       "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": 28467.17273835364,
-      "required_pages_per_second_upper_estimate": 0.3534670119119615
+      "required_token_positions_per_second": 28452.108639258768,
+      "required_pages_per_second_upper_estimate": 0.35327996621749963
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.09455374124217997,
-    "producer_p95_seconds": 0.3116126649983926,
-    "gpu_update_seconds": 0.9712239516764524,
-    "producer_batches_per_second": 10.575996114619151,
-    "gpu_batches_per_second": 1.0296286436036473,
-    "producer_headroom": 10.27166073935522,
+    "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,
     "sustainable": true,
-    "buffer_seconds": 60.215885003940045,
+    "buffer_seconds": 60.24776657975877,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 509643.2155412006,
+    "generation_bytes_per_second": 517929.33889289916,
     "upload_bytes_per_second": 4008350.9631803427,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 109.35757016827596,
+    "estimated_full_shards_per_hour": 111.13557934847098,
     "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-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -213,3 +213,4 @@
 {"step": 136300, "loss": 0.6153386607766151, "lr": 0.0003, "grad_norm": 0.12918039986370972, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19947345253, "tokens_per_second": 150388.36224613513, "data_wait_s": 9.712999599287286e-06, "gpu_peak_gb": 85.46427917480469}
 {"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}

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 sha256:4901cc16c06837b8ac21666ab52b9d488e93b0a3e4c2b17fcd05be9d6042188c
-size 458667
+oid sha256:cb3c7757bc9bc33ad5b464af7ef632930c222981a86c885737cf7179a835c665
+size 460785