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": 77225.96417191433,
-      "required_pages_per_second_upper_estimate": 0.005342671390337328
+      "required_token_positions_per_second": 77255.32623089201,
+      "required_pages_per_second_upper_estimate": 0.005344702725706767
     },
     {
       "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": 9653.24552148929,
-      "required_pages_per_second_upper_estimate": 0.004041612317158346
+      "required_token_positions_per_second": 9656.915778861501,
+      "required_pages_per_second_upper_estimate": 0.0040431489772764
     },
     {
       "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": 19306.49104297858,
-      "required_pages_per_second_upper_estimate": 0.004021032937845022
+      "required_token_positions_per_second": 19313.831557723002,
+      "required_pages_per_second_upper_estimate": 0.004022561773483893
     },
     {
       "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": 28959.736564467872,
-      "required_pages_per_second_upper_estimate": 0.0017360919880824756
+      "required_token_positions_per_second": 28970.7473365845,
+      "required_pages_per_second_upper_estimate": 0.0017367520670583919
     },
     {
       "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": 28959.736564467872,
-      "required_pages_per_second_upper_estimate": 0.0022514921374682077
+      "required_token_positions_per_second": 28970.7473365845,
+      "required_pages_per_second_upper_estimate": 0.002252348176569007
     },
     {
       "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": 28959.736564467872,
-      "required_pages_per_second_upper_estimate": 0.3714644061064875
+      "required_token_positions_per_second": 28970.7473365845,
+      "required_pages_per_second_upper_estimate": 0.3716056404687536
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.36951695220312786,
-    "producer_p95_seconds": 0.21779314000013983,
-    "gpu_update_seconds": 0.954704817098464,
-    "producer_batches_per_second": 2.7062357871210416,
-    "gpu_batches_per_second": 1.0474441755088206,
-    "producer_headroom": 2.5836563421687115,
+    "producer_batch_seconds": 0.05758404768751291,
+    "producer_p95_seconds": 0.19186565700056235,
+    "gpu_update_seconds": 0.9543419670463894,
+    "producer_batches_per_second": 17.365920600556358,
+    "gpu_batches_per_second": 1.0478424239216038,
+    "producer_headroom": 16.573026825506368,
     "sustainable": true,
-    "buffer_seconds": 60.14640347720323,
+    "buffer_seconds": 60.12354392392253,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 315775.4699008209,
+    "generation_bytes_per_second": 340108.25878880336,
     "upload_bytes_per_second": 5407417.345276203,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.75806496399375,
+    "estimated_full_shards_per_hour": 72.9793150210197,
     "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": [

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -78,3 +78,4 @@
 {"step": 767800, "loss": 0.5399699807167053, "lr": 3e-05, "grad_norm": 0.2565081637582412, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112383712276, "tokens_per_second": 152727.7079886464, "data_wait_s": 1.0433000170451123e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 767900, "loss": 0.5667563676834106, "lr": 3e-05, "grad_norm": 0.2591779451913033, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112398361118, "tokens_per_second": 152276.16642086252, "data_wait_s": 1.239199991687201e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 768000, "loss": 0.5218808650970459, "lr": 3e-05, "grad_norm": 0.23926654398524383, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112412781567, "tokens_per_second": 149940.81730008766, "data_wait_s": 1.0976999874401372e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 768100, "loss": 0.5042897462844849, "lr": 3e-05, "grad_norm": 0.2413063474768963, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112427227492, "tokens_per_second": 150296.4566534407, "data_wait_s": 1.1320000339765102e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:88aaab0dca27dd08d00efb91793069dd48815b714dbb7797ef96ce0f2033ce15
-size 170388
+oid sha256:c475c0d3743eb862d778df2ea686d8accb1bb5253263f0642e513321a099c13a
+size 172506