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": 77194.93676858602,
-      "required_pages_per_second_upper_estimate": 0.005340524842581581
+      "required_token_positions_per_second": 77201.62098354206,
+      "required_pages_per_second_upper_estimate": 0.005340987272081756
     },
     {
       "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": 9649.367096073252,
-      "required_pages_per_second_upper_estimate": 0.004039988501427383
+      "required_token_positions_per_second": 9650.202622942757,
+      "required_pages_per_second_upper_estimate": 0.00404033831907986
     },
     {
       "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": 19298.734192146505,
-      "required_pages_per_second_upper_estimate": 0.004019417390378613
+      "required_token_positions_per_second": 19300.405245885515,
+      "required_pages_per_second_upper_estimate": 0.004019765426803799
     },
     {
       "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": 28948.10128821976,
-      "required_pages_per_second_upper_estimate": 0.0017353944710374397
+      "required_token_positions_per_second": 28950.607868828272,
+      "required_pages_per_second_upper_estimate": 0.0017355447367175929
     },
     {
       "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": 28948.10128821976,
-      "required_pages_per_second_upper_estimate": 0.0022505875459180886
+      "required_token_positions_per_second": 28950.607868828272,
+      "required_pages_per_second_upper_estimate": 0.0022507824215351174
     },
     {
       "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": 28948.10128821976,
-      "required_pages_per_second_upper_estimate": 0.37131516127576303
+      "required_token_positions_per_second": 28950.607868828272,
+      "required_pages_per_second_upper_estimate": 0.37134731300045243
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.11855071492931302,
-    "producer_p95_seconds": 0.360649898997508,
-    "gpu_update_seconds": 0.9550885470768742,
-    "producer_batches_per_second": 8.435208514737845,
-    "gpu_batches_per_second": 1.0470233394176707,
-    "producer_headroom": 8.056371044631446,
+    "producer_batch_seconds": 0.11534706257805283,
+    "producer_p95_seconds": 0.356001396998181,
+    "gpu_update_seconds": 0.9550058542905132,
+    "producer_batches_per_second": 8.669488217988404,
+    "gpu_batches_per_second": 1.0471139998852819,
+    "producer_headroom": 8.279412001881553,
     "sustainable": true,
-    "buffer_seconds": 60.17057846584307,
+    "buffer_seconds": 60.165368820302334,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 307336.5761232067,
-    "upload_bytes_per_second": 5002805.577810708,
+    "generation_bytes_per_second": 341725.07458227786,
+    "upload_bytes_per_second": 5131831.486766642,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.94727480671072,
+    "estimated_full_shards_per_hour": 73.32624605275394,
     "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": 3.413111547000881,
-      "bytes": 7579762,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 1.0360288759984542,
       "bytes": 6839680,
@@ -562,6 +556,12 @@
       "bytes": 862575,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.0102765199990245,
+      "bytes": 6370774,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -407,3 +407,4 @@
 {"step": 437800, "loss": 0.5225988626480103, "lr": 0.0003, "grad_norm": 0.18840691812886684, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64081268836, "tokens_per_second": 151101.5145848583, "data_wait_s": 1.42659991979599e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 437900, "loss": 0.575453519821167, "lr": 0.0003, "grad_norm": 0.18456594881201405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64095892710, "tokens_per_second": 151770.3124992193, "data_wait_s": 8.305003575515002e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 438000, "loss": 0.5736438035964966, "lr": 0.0003, "grad_norm": 0.14985151077977857, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64110620537, "tokens_per_second": 152670.74583348134, "data_wait_s": 2.0515995856840163e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 438100, "loss": 0.556441605091095, "lr": 0.0003, "grad_norm": 0.169631654008228, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64125325344, "tokens_per_second": 152286.5565314529, "data_wait_s": 8.811002771835774e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fd43987ed57160aa67ea8944e27317e4e9576a323584bd6d42abaa4527af8f5b
-size 872691
+oid sha256:cc97cab01588c063ec95b7666fc21562dcbc8ee6df667476dacc08dd00271331
+size 874809