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": 76727.44529779312,
-      "required_pages_per_second_upper_estimate": 0.00530818269790247
+      "required_token_positions_per_second": 76783.1878684633,
+      "required_pages_per_second_upper_estimate": 0.005312039098282016
     },
     {
       "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": 9590.93066222414,
-      "required_pages_per_second_upper_estimate": 0.004015522386866263
+      "required_token_positions_per_second": 9597.898483557912,
+      "required_pages_per_second_upper_estimate": 0.004018439668154058
     },
     {
       "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": 19181.86132444828,
-      "required_pages_per_second_upper_estimate": 0.00399507585418186
+      "required_token_positions_per_second": 19195.796967115824,
+      "required_pages_per_second_upper_estimate": 0.003997978281041898
     },
     {
       "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": 28772.791986672422,
-      "required_pages_per_second_upper_estimate": 0.0017248849460914809
+      "required_token_positions_per_second": 28793.69545067374,
+      "required_pages_per_second_upper_estimate": 0.0017261380768406182
     },
     {
       "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": 28772.791986672422,
-      "required_pages_per_second_upper_estimate": 0.0022369580188269077
+      "required_token_positions_per_second": 28793.69545067374,
+      "required_pages_per_second_upper_estimate": 0.0022385831712083888
     },
     {
       "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": 28772.791986672422,
-      "required_pages_per_second_upper_estimate": 0.36906648178797635
+      "required_token_positions_per_second": 28793.69545067374,
+      "required_pages_per_second_upper_estimate": 0.36933460897979425
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.11443751607825448,
-    "producer_p95_seconds": 0.2371655510032724,
-    "gpu_update_seconds": 0.9609077913887041,
-    "producer_batches_per_second": 8.738393092315825,
-    "gpu_batches_per_second": 1.0406825805364737,
-    "producer_headroom": 8.396790006623506,
+    "producer_batch_seconds": 0.07484092403123555,
+    "producer_p95_seconds": 0.21612445100254263,
+    "gpu_update_seconds": 0.9602101976581499,
+    "producer_batches_per_second": 13.361673615663014,
+    "gpu_batches_per_second": 1.041438637538836,
+    "producer_headroom": 12.830015263539469,
     "sustainable": true,
-    "buffer_seconds": 60.537190857488355,
+    "buffer_seconds": 60.49324245246344,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 410871.21346063394,
-    "upload_bytes_per_second": 5346232.266165973,
+    "generation_bytes_per_second": 428974.5715511819,
+    "upload_bytes_per_second": 5209704.617432161,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.16339781631721,
+    "estimated_full_shards_per_hour": 92.0479570379409,
     "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": 1.1759565290012688,
-      "bytes": 13132586,
-      "pages": 31,
-      "shards": 4
-    },
     {
       "seconds": 1.004212905998429,
       "bytes": 6157452,
@@ -562,6 +556,12 @@
       "bytes": 12076168,
       "pages": 35,
       "shards": 5
+    },
+    {
+      "seconds": 1.0162399760010885,
+      "bytes": 1245993,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -180,3 +180,4 @@
 {"step": 622400, "loss": 0.5498073697090149, "lr": 0.0001, "grad_norm": 0.17699649311324087, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91102546752, "tokens_per_second": 150639.0478257129, "data_wait_s": 1.3329998182598501e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 622500, "loss": 0.5145138502120972, "lr": 0.0001, "grad_norm": 0.16401060052470887, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91117207325, "tokens_per_second": 151897.75427026316, "data_wait_s": 1.0744002793217078e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 622600, "loss": 0.5647789835929871, "lr": 0.0001, "grad_norm": 0.16720628825279638, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91131949369, "tokens_per_second": 152741.46973627643, "data_wait_s": 9.515999408904463e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 622700, "loss": 0.5370277762413025, "lr": 0.0001, "grad_norm": 0.14730485395166398, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91146648237, "tokens_per_second": 153021.0069713086, "data_wait_s": 1.3280998246045783e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ac188fc5c35a3046e369ac59be15b9d266a9466684a7169ff873297aeb515df2
-size 388773
+oid sha256:a25cace53d8ed7bfdb4ab1f5a6b24cc0fc595d48473bb7862022a753d7f85acd
+size 390891