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": 76848.11733691524,
-      "required_pages_per_second_upper_estimate": 0.005316531069566644
+      "required_token_positions_per_second": 76499.32631460873,
+      "required_pages_per_second_upper_estimate": 0.005292400897336807
     },
     {
       "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": 9606.014667114405,
-      "required_pages_per_second_upper_estimate": 0.0040218377447239755
+      "required_token_positions_per_second": 9562.41578932609,
+      "required_pages_per_second_upper_estimate": 0.0040035838050421066
     },
     {
       "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": 19212.02933422881,
-      "required_pages_per_second_upper_estimate": 0.004001359055035225
+      "required_token_positions_per_second": 19124.83157865218,
+      "required_pages_per_second_upper_estimate": 0.003983198062108067
     },
     {
       "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": 28818.044001343216,
-      "required_pages_per_second_upper_estimate": 0.0017275977352751693
+      "required_token_positions_per_second": 28687.247367978274,
+      "required_pages_per_second_upper_estimate": 0.0017197566768198363
     },
     {
       "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": 28818.044001343216,
-      "required_pages_per_second_upper_estimate": 0.002240476163924985
+      "required_token_positions_per_second": 28687.247367978274,
+      "required_pages_per_second_upper_estimate": 0.00223030730099445
     },
     {
       "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": 28818.044001343216,
-      "required_pages_per_second_upper_estimate": 0.36964692604434546
+      "required_token_positions_per_second": 28687.247367978274,
+      "required_pages_per_second_upper_estimate": 0.36796920727002314
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.1219470694921938,
-    "producer_p95_seconds": 0.336407959999633,
-    "gpu_update_seconds": 0.9593989098882395,
-    "producer_batches_per_second": 8.20027905684124,
-    "gpu_batches_per_second": 1.0423192998171011,
-    "producer_headroom": 7.867338787912845,
+    "producer_batch_seconds": 0.1137601242578512,
+    "producer_p95_seconds": 0.33080791999964276,
+    "gpu_update_seconds": 0.9637731932015785,
+    "producer_batches_per_second": 8.790426404012868,
+    "gpu_batches_per_second": 1.0375885188070846,
+    "producer_headroom": 8.471977324998951,
     "sustainable": true,
-    "buffer_seconds": 60.44213132295909,
+    "buffer_seconds": 60.71771117169945,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 360481.0273459721,
+    "generation_bytes_per_second": 358753.9760649521,
     "upload_bytes_per_second": 5441238.513238026,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.35083689960834,
+    "estimated_full_shards_per_hour": 76.98025189839765,
     "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-v1/1790868774355361490-de221e1d/train.jsonl

@@ -341,3 +341,4 @@
 {"step": 474300, "loss": 0.5846430659294128, "lr": 0.0003, "grad_norm": 0.15215446976376132, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69422543844, "tokens_per_second": 152182.48746607485, "data_wait_s": 1.353699917672202e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 474400, "loss": 0.6118581891059875, "lr": 0.0003, "grad_norm": 0.16554137706445823, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69437260186, "tokens_per_second": 113874.5264129922, "data_wait_s": 9.49700188357383e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 474500, "loss": 0.5762530565261841, "lr": 0.0003, "grad_norm": 0.17740558733894285, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69451916420, "tokens_per_second": 152573.086350674, "data_wait_s": 1.3256001693662256e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 474600, "loss": 0.5983161330223083, "lr": 0.0003, "grad_norm": 0.1665082044412824, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69466474121, "tokens_per_second": 151087.0543587786, "data_wait_s": 1.1262003681622446e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:67cddde70bd74afc797d621c9995667919c6873cb019e880c7d6f5ea48b8c12c
-size 731337
+oid sha256:6187e18cf4fa7288adf703a2b23c9a61e4212c15a16867b25299b983096064ee
+size 733455