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": 76151.8360304024,
-      "required_pages_per_second_upper_estimate": 0.0052683607131868695
+      "required_token_positions_per_second": 75609.17320586169,
+      "required_pages_per_second_upper_estimate": 0.005230818039833913
     },
     {
       "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": 9518.9795038003,
-      "required_pages_per_second_upper_estimate": 0.003985397939345244
+      "required_token_positions_per_second": 9451.146650732711,
+      "required_pages_per_second_upper_estimate": 0.003956997740276894
     },
     {
       "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": 19037.9590076006,
-      "required_pages_per_second_upper_estimate": 0.003965104796541783
+      "required_token_positions_per_second": 18902.293301465423,
+      "required_pages_per_second_upper_estimate": 0.003936849207699089
     },
     {
       "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": 28556.938511400895,
-      "required_pages_per_second_upper_estimate": 0.0017119448598381256
+      "required_token_positions_per_second": 28353.439952198132,
+      "required_pages_per_second_upper_estimate": 0.0016997454319382284
     },
     {
       "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": 28556.938511400895,
-      "required_pages_per_second_upper_estimate": 0.0022201763605636444
+      "required_token_positions_per_second": 28353.439952198132,
+      "required_pages_per_second_upper_estimate": 0.002204355242681195
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28556.938511400895,
-      "required_pages_per_second_upper_estimate": 0.3662977451725978
+      "required_token_positions_per_second": 28353.439952198132,
+      "required_pages_per_second_upper_estimate": 0.363687484154874
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09642930005475137,
-    "producer_p95_seconds": 0.24687195099977544,
-    "gpu_update_seconds": 0.9681710099617991,
-    "producer_batches_per_second": 10.370292011164784,
-    "gpu_batches_per_second": 1.0328753801866644,
-    "producer_headroom": 10.040216090048185,
+    "producer_batch_seconds": 0.37327320978134537,
+    "producer_p95_seconds": 0.24072316799902183,
+    "gpu_update_seconds": 0.9751197754703679,
+    "producer_batches_per_second": 2.679002869200756,
+    "gpu_batches_per_second": 1.0255150445673513,
+    "producer_headroom": 2.6123486762995127,
     "sustainable": true,
-    "buffer_seconds": 60.02660261763154,
+    "buffer_seconds": 60.45742607916281,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 434355.6684708574,
+    "generation_bytes_per_second": 398124.5243090257,
     "upload_bytes_per_second": 4918447.240584663,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.20261517137804,
+    "estimated_full_shards_per_hour": 85.42825505211906,
     "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

@@ -124,3 +124,4 @@
 {"step": 732400, "loss": 0.5128567814826965, "lr": 5e-05, "grad_norm": 0.21129127978977358, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107202673752, "tokens_per_second": 150691.97426144668, "data_wait_s": 2.23299994104309e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 732500, "loss": 0.5607808828353882, "lr": 5e-05, "grad_norm": 0.2701238554798232, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107217652722, "tokens_per_second": 153406.1708068083, "data_wait_s": 1.7020000086631626e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 732600, "loss": 0.5459712743759155, "lr": 5e-05, "grad_norm": 0.22150784871969925, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107232409461, "tokens_per_second": 150610.48302994028, "data_wait_s": 1.8864000594476238e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 732700, "loss": 0.5102741122245789, "lr": 5e-05, "grad_norm": 0.21796404490415877, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107247016435, "tokens_per_second": 149856.62709858778, "data_wait_s": 1.8905999240814708e-05, "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 sha256:4cee609c0974b0d3577d843031c8afc8a12097722e58e186fbd2ef779f2fffaa
-size 268599
+oid sha256:b2312787e030b7dd611474f85055e2ed781405eab7f6749b5c537aa91b8733c9
+size 270717