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": 75296.62354967698,
-      "required_pages_per_second_upper_estimate": 0.005209195129403938
+      "required_token_positions_per_second": 75785.46714867918,
+      "required_pages_per_second_upper_estimate": 0.005243014463856332
     },
     {
       "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": 9412.077943709623,
-      "required_pages_per_second_upper_estimate": 0.0039406404884937025
+      "required_token_positions_per_second": 9473.183393584897,
+      "required_pages_per_second_upper_estimate": 0.003966224064329585
     },
     {
       "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": 18824.155887419245,
-      "required_pages_per_second_upper_estimate": 0.00392057524497545
+      "required_token_positions_per_second": 18946.366787169794,
+      "required_pages_per_second_upper_estimate": 0.003946028552475331
     },
     {
       "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": 28236.233831128866,
-      "required_pages_per_second_upper_estimate": 0.0016927191039435103
+      "required_token_positions_per_second": 28419.550180754693,
+      "required_pages_per_second_upper_estimate": 0.0017037086391957202
     },
     {
       "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": 28236.233831128866,
-      "required_pages_per_second_upper_estimate": 0.0021952429823033025
+      "required_token_positions_per_second": 28419.550180754693,
+      "required_pages_per_second_upper_estimate": 0.0022094950221633034
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28236.233831128866,
-      "required_pages_per_second_upper_estimate": 0.3621840898799254
+      "required_token_positions_per_second": 28419.550180754693,
+      "required_pages_per_second_upper_estimate": 0.3645354751831646
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.10830349731230626,
-    "producer_p95_seconds": 0.332312954000372,
-    "gpu_update_seconds": 0.9791674118210351,
-    "producer_batches_per_second": 9.233312171964112,
-    "gpu_batches_per_second": 1.021275818544881,
-    "producer_headroom": 9.04095838195776,
+    "producer_batch_seconds": 0.14315138336684186,
+    "producer_p95_seconds": 0.3367123209973215,
+    "gpu_update_seconds": 0.972851428828132,
+    "producer_batches_per_second": 6.985611850060751,
+    "gpu_batches_per_second": 1.02790618419975,
+    "producer_headroom": 6.795962469570333,
     "sustainable": true,
-    "buffer_seconds": 60.708379532904175,
+    "buffer_seconds": 60.316788587344185,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 161,
+    "upload_pages_per_commit": 164,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 644138.1665715517,
-    "upload_bytes_per_second": 3812793.992886595,
+    "generation_bytes_per_second": 616272.5528772353,
+    "upload_bytes_per_second": 3821791.9218810773,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.10852788858372,
+    "estimated_full_shards_per_hour": 66.11887187832734,
     "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": 1.6743759019991558,
-      "bytes": 6275115,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 2.272818253000878,
       "bytes": 7589463,
@@ -562,6 +556,12 @@
       "bytes": 942860,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.5147400999994716,
+      "bytes": 6664610,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -351,3 +351,4 @@
 {"step": 835100, "loss": 0.5472451448440552, "lr": 3e-05, "grad_norm": 0.2910525719190749, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122234886213, "tokens_per_second": 139815.18583565488, "data_wait_s": 1.1014002666343004e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 835200, "loss": 0.5077463388442993, "lr": 3e-05, "grad_norm": 0.2840245954475329, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122249570745, "tokens_per_second": 151653.3067558072, "data_wait_s": 9.439994755666703e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 835300, "loss": 0.5298817157745361, "lr": 3e-05, "grad_norm": 0.29290207557855097, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122264067910, "tokens_per_second": 150097.1598721705, "data_wait_s": 1.545899431221187e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 835400, "loss": 0.5379299521446228, "lr": 3e-05, "grad_norm": 0.305677779022904, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122278756085, "tokens_per_second": 152780.8836670833, "data_wait_s": 1.3860997569281608e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8097f1dc78c6f934a6a245b21b16a0a6ec2a06d2199a66944108747cf465f468
-size 753300
+oid sha256:21eb83f433efdce6c31a93367fb7e942ce8811844918eb1c307cc62fae6c33c8
+size 755418