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": 77046.67721046576,
-      "required_pages_per_second_upper_estimate": 0.005330267902341252
+      "required_token_positions_per_second": 76842.31754405703,
+      "required_pages_per_second_upper_estimate": 0.005316129826439341
     },
     {
       "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": 9630.83465130822,
-      "required_pages_per_second_upper_estimate": 0.004032229353805717
+      "required_token_positions_per_second": 9605.28969300713,
+      "required_pages_per_second_upper_estimate": 0.0040215342132044396
     },
     {
       "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": 19261.66930261644,
-      "required_pages_per_second_upper_estimate": 0.004011697751356369
+      "required_token_positions_per_second": 19210.57938601426,
+      "required_pages_per_second_upper_estimate": 0.00400105706905984
     },
     {
       "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": 28892.503953924657,
-      "required_pages_per_second_upper_estimate": 0.0017320614957386622
+      "required_token_positions_per_second": 28815.869079021388,
+      "required_pages_per_second_upper_estimate": 0.0017274673520028898
     },
     {
       "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": 28892.503953924657,
-      "required_pages_per_second_upper_estimate": 0.0022462650977234733
+      "required_token_positions_per_second": 28815.869079021388,
+      "required_pages_per_second_upper_estimate": 0.002240307073280935
     },
     {
       "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": 28892.503953924657,
-      "required_pages_per_second_upper_estimate": 0.37060201836719203
+      "required_token_positions_per_second": 28815.869079021388,
+      "required_pages_per_second_upper_estimate": 0.3696190284760507
     }
   ],
   "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": 0.04745994811718646,
-    "producer_p95_seconds": 0.16071723899949575,
-    "gpu_update_seconds": 0.956926407073984,
-    "producer_batches_per_second": 21.070398086631588,
-    "gpu_batches_per_second": 1.0450124404631314,
-    "producer_headroom": 20.162820336658914,
+    "producer_batch_seconds": 0.07036065811724512,
+    "producer_p95_seconds": 0.16610220699931233,
+    "gpu_update_seconds": 0.9594713220059837,
+    "producer_batches_per_second": 14.212487869764594,
+    "gpu_batches_per_second": 1.0422406350919193,
+    "producer_headroom": 13.636474525397043,
     "sustainable": true,
-    "buffer_seconds": 60.28636364566099,
+    "buffer_seconds": 60.44669328637698,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 373441.0798832574,
+    "generation_bytes_per_second": 371684.31311117695,
     "upload_bytes_per_second": 5004604.73724531,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.1317624795274,
+    "estimated_full_shards_per_hour": 79.75480122567636,
     "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-v1/1791399463488568880-054ba646/train.jsonl

@@ -42,3 +42,4 @@
 {"step": 889200, "loss": 0.5401551127433777, "lr": 3e-05, "grad_norm": 0.31936998721257415, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130154458253, "tokens_per_second": 153469.32233748038, "data_wait_s": 9.684000360721257e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 889300, "loss": 0.5247880816459656, "lr": 3e-05, "grad_norm": 0.30831380885092613, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130169050578, "tokens_per_second": 152004.8896396374, "data_wait_s": 1.3217999367043376e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 889400, "loss": 0.5499499440193176, "lr": 3e-05, "grad_norm": 0.32768441044821467, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130183539947, "tokens_per_second": 150971.01519701548, "data_wait_s": 1.8912000086857006e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 889500, "loss": 0.5187528729438782, "lr": 3e-05, "grad_norm": 0.30286175324321074, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130197951774, "tokens_per_second": 143902.7250845003, "data_wait_s": 2.0712999685201794e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0320acff8481d1d83f8d603ff88709f8f205906c3bee10899db4218508e5afcc
-size 93357
+oid sha256:41a49eb04beee8d71949feeededf9476570ec0dd3be6376642922faf3fdf284d
+size 95475