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": 76767.06792184967,
-      "required_pages_per_second_upper_estimate": 0.005310923882971856
+      "required_token_positions_per_second": 77070.86913759257,
+      "required_pages_per_second_upper_estimate": 0.005331941555992889
     },
     {
       "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": 9595.883490231208,
-      "required_pages_per_second_upper_estimate": 0.004017596032526011
+      "required_token_positions_per_second": 9633.858642199071,
+      "required_pages_per_second_upper_estimate": 0.004033495435643607
     },
     {
       "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": 19191.766980462417,
-      "required_pages_per_second_upper_estimate": 0.003997138941099888
+      "required_token_positions_per_second": 19267.717284398143,
+      "required_pages_per_second_upper_estimate": 0.004012957386465493
     },
     {
       "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": 28787.65047069362,
-      "required_pages_per_second_upper_estimate": 0.0017257756895209602
+      "required_token_positions_per_second": 28901.575926597216,
+      "required_pages_per_second_upper_estimate": 0.0017326053466482818
     },
     {
       "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": 28787.65047069362,
-      "required_pages_per_second_upper_estimate": 0.0022381132006039917
+      "required_token_positions_per_second": 28901.575926597216,
+      "required_pages_per_second_upper_estimate": 0.002246970403695374
     },
     {
       "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": 28787.65047069362,
-      "required_pages_per_second_upper_estimate": 0.36925707046720313
+      "required_token_positions_per_second": 28901.575926597216,
+      "required_pages_per_second_upper_estimate": 0.37071838389191025
     }
   ],
   "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": 0.15570629900776112,
-    "producer_p95_seconds": 0.35151863299688557,
-    "gpu_update_seconds": 0.9604118275698182,
-    "producer_batches_per_second": 6.422347755823003,
-    "gpu_batches_per_second": 1.0412199967698792,
-    "producer_headroom": 6.168098745458891,
+    "producer_batch_seconds": 0.09680660765639004,
+    "producer_p95_seconds": 0.3401595120012644,
+    "gpu_update_seconds": 0.9566260355566428,
+    "producer_batches_per_second": 10.329873385806962,
+    "gpu_batches_per_second": 1.0453405644747256,
+    "producer_headroom": 9.88182582486659,
     "sustainable": true,
-    "buffer_seconds": 60.505945136898546,
+    "buffer_seconds": 60.267440240068495,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 324783.5938401433,
+    "generation_bytes_per_second": 324576.0783629743,
     "upload_bytes_per_second": 4574155.459772348,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.69099866297935,
+    "estimated_full_shards_per_hour": 69.64647067229195,
     "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-v1/1791039222312944374-63c10d34/train.jsonl

@@ -413,3 +413,4 @@
 {"step": 645700, "loss": 0.5570340156555176, "lr": 0.0001, "grad_norm": 0.17565000479487577, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94513673484, "tokens_per_second": 150995.53033105843, "data_wait_s": 8.601004083175212e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 645800, "loss": 0.5187062621116638, "lr": 0.0001, "grad_norm": 0.17557828594077124, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94528455980, "tokens_per_second": 141509.37739036925, "data_wait_s": 1.266899926122278e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 645900, "loss": 0.5465184450149536, "lr": 0.0001, "grad_norm": 0.1900253611540677, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94543151532, "tokens_per_second": 152534.39588219472, "data_wait_s": 1.4863995602354407e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 646000, "loss": 0.6285852789878845, "lr": 0.0001, "grad_norm": 0.19419523950274928, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94557920764, "tokens_per_second": 152505.10645218327, "data_wait_s": 1.3965996913611889e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9f1daf6e49b166c2f1f34d72b6f3d957289b5a9721fbca66caec9f6625f42462
-size 886182
+oid sha256:0ca22e66e997a6d374979b03f33e6bd78b4a821b29229f1f90634f98ace598a6
+size 888300