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": 77334.66922500326,
-      "required_pages_per_second_upper_estimate": 0.005350191858140494
+      "required_token_positions_per_second": 77463.90182225917,
+      "required_pages_per_second_upper_estimate": 0.005359132469079596
     },
     {
       "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": 9666.833653125408,
-      "required_pages_per_second_upper_estimate": 0.004047301384121933
+      "required_token_positions_per_second": 9682.987727782396,
+      "required_pages_per_second_upper_estimate": 0.0040540647578453754
     },
     {
       "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": 19333.667306250816,
-      "required_pages_per_second_upper_estimate": 0.004026693036798371
+      "required_token_positions_per_second": 19365.975455564792,
+      "required_pages_per_second_upper_estimate": 0.004033421972277279
     },
     {
       "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": 29000.500959376222,
-      "required_pages_per_second_upper_estimate": 0.0017385357513136042
+      "required_token_positions_per_second": 29048.96318334719,
+      "required_pages_per_second_upper_estimate": 0.0017414409876431323
     },
     {
       "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": 29000.500959376222,
-      "required_pages_per_second_upper_estimate": 0.002254661389868708
+      "required_token_positions_per_second": 29048.96318334719,
+      "required_pages_per_second_upper_estimate": 0.0022584291146196555
     },
     {
       "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": 29000.500959376222,
-      "required_pages_per_second_upper_estimate": 0.3719872879949747
+      "required_token_positions_per_second": 29048.96318334719,
+      "required_pages_per_second_upper_estimate": 0.3726089093693923
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,20 +155,20 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.03864659044531038,
-    "producer_p95_seconds": 0.1129381639998428,
-    "gpu_update_seconds": 0.9533628415153655,
-    "producer_batches_per_second": 25.875503853700145,
-    "gpu_batches_per_second": 1.0489185821533644,
-    "producer_headroom": 24.668743879605362,
+    "producer_batch_seconds": 0.05257124209375519,
+    "producer_p95_seconds": 0.1284731999999167,
+    "gpu_update_seconds": 0.9517723515808538,
+    "producer_batches_per_second": 19.02180660324911,
+    "gpu_batches_per_second": 1.0506714114347218,
+    "producer_headroom": 18.10442960209062,
     "sustainable": true,
-    "buffer_seconds": 60.06185901546803,
+    "buffer_seconds": 60.913430501174645,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 68,
+    "upload_pages_per_commit": 63,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -9,3 +9,4 @@
 {"step": 525900, "loss": 0.5797033309936523, "lr": 0.0003, "grad_norm": 0.16494069511746742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76974555081, "tokens_per_second": 155318.69931265115, "data_wait_s": 1.4172000192047562e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 526000, "loss": 0.582292377948761, "lr": 0.0003, "grad_norm": 0.17085759702452008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76989107617, "tokens_per_second": 152493.29559283511, "data_wait_s": 1.432899989595171e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 526100, "loss": 0.6050062775611877, "lr": 0.0003, "grad_norm": 0.15781412739970652, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77003576077, "tokens_per_second": 151795.13748489312, "data_wait_s": 1.161499994850601e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 526200, "loss": 0.5432044863700867, "lr": 0.0003, "grad_norm": 0.1561291186653021, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77018180283, "tokens_per_second": 153366.29446798816, "data_wait_s": 1.1892000202351483e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cdaa50810b536ad6340762a2c0a538fc971a0d2af2df476ea472e6a94f91aade
-size 23463
+oid sha256:66dc3bfa29959c23babf75712635cdbd690bbdfc71e4fb716fb08d54e8fd832b
+size 25581