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": 75775.32757387195,
-      "required_pages_per_second_upper_estimate": 0.006568013343291299
+      "required_token_positions_per_second": 75906.71047456573,
+      "required_pages_per_second_upper_estimate": 0.006579401280136518
     },
     {
       "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": 9471.915946733994,
-      "required_pages_per_second_upper_estimate": 0.003778488888915747
+      "required_token_positions_per_second": 9488.338809320716,
+      "required_pages_per_second_upper_estimate": 0.003785040214345267
     },
     {
       "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": 18943.83189346799,
-      "required_pages_per_second_upper_estimate": 0.004382322108499519
+      "required_token_positions_per_second": 18976.67761864143,
+      "required_pages_per_second_upper_estimate": 0.004389920388953376
     },
     {
       "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": 28415.74784020198,
-      "required_pages_per_second_upper_estimate": 0.001730369957561349
+      "required_token_positions_per_second": 28465.01642796215,
+      "required_pages_per_second_upper_estimate": 0.0017333701560636407
     },
     {
       "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": 28415.74784020198,
-      "required_pages_per_second_upper_estimate": 0.0021875229285632227
+      "required_token_positions_per_second": 28465.01642796215,
+      "required_pages_per_second_upper_estimate": 0.0021913157608331804
     },
     {
       "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": 28415.74784020198,
-      "required_pages_per_second_upper_estimate": 0.3528284867849806
+      "required_token_positions_per_second": 28465.01642796215,
+      "required_pages_per_second_upper_estimate": 0.3534402377536058
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.30456499728919084,
-    "producer_p95_seconds": 0.31014185399908456,
-    "gpu_update_seconds": 0.9729816070821199,
-    "producer_batches_per_second": 3.2833713949422725,
-    "gpu_batches_per_second": 1.027768657414713,
-    "producer_headroom": 3.1946599764983943,
+    "producer_batch_seconds": 0.09441465291425288,
+    "producer_p95_seconds": 0.29237967700100853,
+    "gpu_update_seconds": 0.9712975248045328,
+    "producer_batches_per_second": 10.591576298100646,
+    "gpu_batches_per_second": 1.029550652053029,
+    "producer_headroom": 10.287571842123514,
     "sustainable": true,
-    "buffer_seconds": 60.324859639091436,
+    "buffer_seconds": 60.22044653788103,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 395786.1447691166,
-    "upload_bytes_per_second": 4065002.8724211757,
+    "generation_bytes_per_second": 380921.4138313177,
+    "upload_bytes_per_second": 4062323.194923724,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.92649323754428,
+    "estimated_full_shards_per_hour": 81.73686801151894,
     "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": 1.2741842929972336,
-      "bytes": 12412930,
-      "pages": 39,
-      "shards": 4
-    },
     {
       "seconds": 1.0742992420018709,
       "bytes": 6000047,
@@ -562,6 +556,12 @@
       "bytes": 5888838,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.2323054039989074,
+      "bytes": 11963716,
+      "pages": 40,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -276,3 +276,4 @@
 {"step": 342600, "loss": 0.5854822285473347, "lr": 0.0003, "grad_norm": 0.1590529946437544, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50147690603, "tokens_per_second": 150395.6236343508, "data_wait_s": 1.6707999748177826e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 342700, "loss": 0.575041014701128, "lr": 0.0003, "grad_norm": 0.1384718057305531, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50162281932, "tokens_per_second": 148025.59805104564, "data_wait_s": 1.4633998944191262e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 342800, "loss": 0.5893674921244383, "lr": 0.0003, "grad_norm": 0.15724862747313545, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50176985482, "tokens_per_second": 150103.52280996006, "data_wait_s": 1.168600283563137e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 342900, "loss": 0.6649965941905975, "lr": 0.0003, "grad_norm": 0.1878389021126051, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50191531985, "tokens_per_second": 147458.1379250797, "data_wait_s": 1.1872001778101549e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cdd95dc18c9971f618e2a17f8f26632c2930c454997958b9204a6a9e34312f3f
-size 592884
+oid sha256:e4a93674ae258f73344918e82589df072580e3769006c1fa1286836e8547a08c
+size 595002