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": 75164.68315444687,
-      "required_pages_per_second_upper_estimate": 0.006515084232679665
+      "required_token_positions_per_second": 75150.54189164305,
+      "required_pages_per_second_upper_estimate": 0.006513858503860532
     },
     {
       "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": 9395.58539430586,
-      "required_pages_per_second_upper_estimate": 0.003748039490308704
+      "required_token_positions_per_second": 9393.817736455381,
+      "required_pages_per_second_upper_estimate": 0.0037473343451633083
     },
     {
       "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": 18791.17078861172,
-      "required_pages_per_second_upper_estimate": 0.004347006648634704
+      "required_token_positions_per_second": 18787.635472910762,
+      "required_pages_per_second_upper_estimate": 0.004346188815566718
     },
     {
       "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": 28186.756182917576,
-      "required_pages_per_second_upper_estimate": 0.0017164255670591034
+      "required_token_positions_per_second": 28181.453209366144,
+      "required_pages_per_second_upper_estimate": 0.0017161026437923733
     },
     {
       "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": 28186.756182917576,
-      "required_pages_per_second_upper_estimate": 0.002169894516895991
+      "required_token_positions_per_second": 28181.453209366144,
+      "required_pages_per_second_upper_estimate": 0.0021694862793124317
     },
     {
       "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": 28186.756182917576,
-      "required_pages_per_second_upper_estimate": 0.349985176787285
+      "required_token_positions_per_second": 28181.453209366144,
+      "required_pages_per_second_upper_estimate": 0.34991933160368704
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.085689027265623,
-    "producer_p95_seconds": 0.1888412669995887,
-    "gpu_update_seconds": 0.9808861942317404,
-    "producer_batches_per_second": 11.67010563558099,
-    "gpu_batches_per_second": 1.0194862624029795,
-    "producer_headroom": 11.447045503167423,
+    "producer_batch_seconds": 0.0709647496796535,
+    "producer_p95_seconds": 0.18526915700022073,
+    "gpu_update_seconds": 0.981070770005968,
+    "producer_batches_per_second": 14.091503239483883,
+    "gpu_batches_per_second": 1.0192944592508009,
+    "producer_headroom": 13.824761933702044,
     "sustainable": true,
-    "buffer_seconds": 60.814944042367905,
+    "buffer_seconds": 60.826387740370016,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 394124.7278405155,
-    "upload_bytes_per_second": 3044829.783041059,
+    "generation_bytes_per_second": 394478.5167755432,
+    "upload_bytes_per_second": 3078760.5405726614,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.56999184047316,
+    "estimated_full_shards_per_hour": 84.64590671014521,
     "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": [
@@ -424,6 +424,12 @@
       "bytes": 11777166,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 1.0721420089994353,
+      "bytes": 6211390,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -48,3 +48,4 @@
 {"step": 241900, "loss": 0.6098959427326918, "lr": 0.0003, "grad_norm": 0.14584189645100581, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35406558433, "tokens_per_second": 147494.01958781225, "data_wait_s": 1.374099974782439e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 242000, "loss": 0.6067950278520584, "lr": 0.0003, "grad_norm": 0.14927075253257158, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35421246671, "tokens_per_second": 149102.67931153878, "data_wait_s": 1.817799966374878e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 242100, "loss": 0.6696290075778961, "lr": 0.0003, "grad_norm": 0.17687013251592487, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35435978706, "tokens_per_second": 149735.172557017, "data_wait_s": 1.2118000086047687e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 242200, "loss": 0.6197372898459435, "lr": 0.0003, "grad_norm": 0.1444105900824176, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35450713692, "tokens_per_second": 150344.98341179563, "data_wait_s": 1.4904000636306591e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:57406610cdc752c4c4064cd12a8cf27a8f11f2f4fc50b44116ab5da79e9fd049
-size 106848
+oid sha256:bd603af2fb6cf5f44d0bfaa62f4b78403cd0900af79f192c05c9ff0ce6256a65
+size 108966