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": 76984.2191212589,
-      "required_pages_per_second_upper_estimate": 0.005325946906807187
+      "required_token_positions_per_second": 76872.66985732946,
+      "required_pages_per_second_upper_estimate": 0.0053182296698985965
     },
     {
       "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": 9623.027390157362,
-      "required_pages_per_second_upper_estimate": 0.004028960616596006
+      "required_token_positions_per_second": 9609.083732166182,
+      "required_pages_per_second_upper_estimate": 0.004023122698171788
     },
     {
       "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": 19246.054780314724,
-      "required_pages_per_second_upper_estimate": 0.004008445658143566
+      "required_token_positions_per_second": 19218.167464332364,
+      "required_pages_per_second_upper_estimate": 0.004002637465662419
     },
     {
       "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": 28869.082170472084,
-      "required_pages_per_second_upper_estimate": 0.001730657395583673
+      "required_token_positions_per_second": 28827.251196498542,
+      "required_pages_per_second_upper_estimate": 0.001728149692566158
     },
     {
       "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": 28869.082170472084,
-      "required_pages_per_second_upper_estimate": 0.0022444441570815734
+      "required_token_positions_per_second": 28827.251196498542,
+      "required_pages_per_second_upper_estimate": 0.00224119198285014
     },
     {
       "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": 28869.082170472084,
-      "required_pages_per_second_upper_estimate": 0.37030158887741416
+      "required_token_positions_per_second": 28827.251196498542,
+      "required_pages_per_second_upper_estimate": 0.36976502605788203
     }
   ],
   "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": 0.18400648166411315,
-    "producer_p95_seconds": 0.26011092200133135,
-    "gpu_update_seconds": 0.9577027713156384,
-    "producer_batches_per_second": 5.434591167420981,
-    "gpu_batches_per_second": 1.044165298411172,
-    "producer_headroom": 5.204723022006564,
+    "producer_batch_seconds": 0.10475422732815787,
+    "producer_p95_seconds": 0.25452649900034885,
+    "gpu_update_seconds": 0.9590924854936644,
+    "producer_batches_per_second": 9.546154131492512,
+    "gpu_batches_per_second": 1.0426523146881708,
+    "producer_headroom": 9.155644692878766,
     "sustainable": true,
-    "buffer_seconds": 60.33527459288522,
+    "buffer_seconds": 60.42282658610086,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 330756.23118513095,
+    "generation_bytes_per_second": 317299.45236289623,
     "upload_bytes_per_second": 3612608.3228906672,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.97258760133215,
+    "estimated_full_shards_per_hour": 68.08507612385907,
     "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-v1/1791342976705245012-6808dd50/train.jsonl

@@ -215,3 +215,4 @@
 {"step": 864000, "loss": 0.5007447004318237, "lr": 3e-05, "grad_norm": 0.3006990500875119, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126465423623, "tokens_per_second": 151092.7277189139, "data_wait_s": 9.35500065679662e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 864100, "loss": 0.5419777631759644, "lr": 3e-05, "grad_norm": 0.31882175322727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126479984323, "tokens_per_second": 150390.26237038162, "data_wait_s": 1.6076999600045383e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 864200, "loss": 0.5264800786972046, "lr": 3e-05, "grad_norm": 0.31084709395988436, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126494659108, "tokens_per_second": 152552.89609263383, "data_wait_s": 1.3656001101480797e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 864300, "loss": 0.5039174556732178, "lr": 3e-05, "grad_norm": 0.2950963750173123, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126509294726, "tokens_per_second": 151479.05976698076, "data_wait_s": 1.2817003153031692e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bffc199edc06fd63bcb49f504078661ade2dc48b886ea24ace9d3fd1be9170ec
-size 462903
+oid sha256:a38226c65c28f3f0e6da591bc3c161b2e0fe0483b79e9bf3e1e1f25e838e6488
+size 465021