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": 76567.91338535276,
-      "required_pages_per_second_upper_estimate": 0.00529714591003481
+      "required_token_positions_per_second": 76608.32019927236,
+      "required_pages_per_second_upper_estimate": 0.005299941347178493
     },
     {
       "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": 9570.989173169095,
-      "required_pages_per_second_upper_estimate": 0.004007173301824559
+      "required_token_positions_per_second": 9576.040024909045,
+      "required_pages_per_second_upper_estimate": 0.004009287987974299
     },
     {
       "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": 19141.97834633819,
-      "required_pages_per_second_upper_estimate": 0.00398676928162639
+      "required_token_positions_per_second": 19152.08004981809,
+      "required_pages_per_second_upper_estimate": 0.003988873200061418
     },
     {
       "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": 28712.96751950728,
-      "required_pages_per_second_upper_estimate": 0.0017212985606316028
+      "required_token_positions_per_second": 28728.120074727136,
+      "required_pages_per_second_upper_estimate": 0.001722206933180421
     },
     {
       "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": 28712.96751950728,
-      "required_pages_per_second_upper_estimate": 0.0022323069296448397
+      "required_token_positions_per_second": 28728.120074727136,
+      "required_pages_per_second_upper_estimate": 0.0022334849741641365
     },
     {
       "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": 28712.96751950728,
-      "required_pages_per_second_upper_estimate": 0.36829911775769014
+      "required_token_positions_per_second": 28728.120074727136,
+      "required_pages_per_second_upper_estimate": 0.3684934784665042
     }
   ],
   "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": 0.1062490240152556,
-    "producer_p95_seconds": 0.3072356039992883,
-    "gpu_update_seconds": 0.962909876215903,
-    "producer_batches_per_second": 9.411851160688466,
-    "gpu_batches_per_second": 1.0385187904914381,
-    "producer_headroom": 9.062764436101032,
+    "producer_batch_seconds": 0.5142956878435712,
+    "producer_p95_seconds": 0.332353884994518,
+    "gpu_update_seconds": 0.9624019924757504,
+    "producer_batches_per_second": 1.9444067365078925,
+    "gpu_batches_per_second": 1.0390668429805823,
+    "producer_headroom": 1.8713009173984672,
     "sustainable": true,
-    "buffer_seconds": 60.663322201601886,
+    "buffer_seconds": 60.63132552597228,
     "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": 293925.11613120016,
+    "generation_bytes_per_second": 270219.15426799917,
     "upload_bytes_per_second": 4879999.152468167,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.06948769523624,
+    "estimated_full_shards_per_hour": 57.98274012594205,
     "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-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -379,3 +379,4 @@
 {"step": 757900, "loss": 0.5424357652664185, "lr": 5e-05, "grad_norm": 0.23454591368320088, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110934907123, "tokens_per_second": 151033.32091534746, "data_wait_s": 1.4106000890024006e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 758000, "loss": 0.5558410286903381, "lr": 5e-05, "grad_norm": 0.257306425486549, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110949712058, "tokens_per_second": 152647.6719865972, "data_wait_s": 1.3356999261304736e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 758100, "loss": 0.5570091605186462, "lr": 5e-05, "grad_norm": 0.24113663297598134, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110964444119, "tokens_per_second": 152024.26559549387, "data_wait_s": 1.5047000488266349e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 758200, "loss": 0.5575207471847534, "lr": 5e-05, "grad_norm": 0.24727222334973173, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110979052216, "tokens_per_second": 151033.8948268855, "data_wait_s": 9.016999683808535e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d0d6ad259ea1d77c76f0858079d90389089da5b6a77f2d3769acdf37a9143a69
-size 812604
+oid sha256:d936f517576b8c77450b4db7db3764b12a9c5e6a74e98af3de56828b018d51cc
+size 814722