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": 76818.21459025308,
-      "required_pages_per_second_upper_estimate": 0.005314462328168625
+      "required_token_positions_per_second": 76811.89325333189,
+      "required_pages_per_second_upper_estimate": 0.005314025003412894
     },
     {
       "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": 9602.276823781634,
-      "required_pages_per_second_upper_estimate": 0.004020272787775589
+      "required_token_positions_per_second": 9601.486656666486,
+      "required_pages_per_second_upper_estimate": 0.004019941961305042
     },
     {
       "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": 19204.55364756327,
-      "required_pages_per_second_upper_estimate": 0.003999802066649908
+      "required_token_positions_per_second": 19202.973313332972,
+      "required_pages_per_second_upper_estimate": 0.003999472924705954
     },
     {
       "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": 28806.830471344903,
-      "required_pages_per_second_upper_estimate": 0.0017269255012738393
+      "required_token_positions_per_second": 28804.45996999946,
+      "required_pages_per_second_upper_estimate": 0.001726783393337726
     },
     {
       "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": 28806.830471344903,
-      "required_pages_per_second_upper_estimate": 0.002239604361290722
+      "required_token_positions_per_second": 28804.45996999946,
+      "required_pages_per_second_upper_estimate": 0.002239420065237847
     },
     {
       "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": 28806.830471344903,
-      "required_pages_per_second_upper_estimate": 0.36950309092167755
+      "required_token_positions_per_second": 28804.45996999946,
+      "required_pages_per_second_upper_estimate": 0.36947268467566424
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10314217333585418,
-    "producer_p95_seconds": 0.20737120300145762,
-    "gpu_update_seconds": 0.9597723716082673,
-    "producer_batches_per_second": 9.695355136097186,
-    "gpu_batches_per_second": 1.0419137178582503,
-    "producer_headroom": 9.305333992556392,
+    "producer_batch_seconds": 0.058074257171838894,
+    "producer_p95_seconds": 0.2007301150006242,
+    "gpu_update_seconds": 0.9598513573521621,
+    "producer_batches_per_second": 17.21933346544664,
+    "gpu_batches_per_second": 1.041827979238985,
+    "producer_headroom": 16.528000599508466,
     "sustainable": true,
-    "buffer_seconds": 60.46565941132084,
+    "buffer_seconds": 60.47063551318621,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 365888.4023332574,
-    "upload_bytes_per_second": 5279063.544478442,
+    "generation_bytes_per_second": 350117.6446957247,
+    "upload_bytes_per_second": 5292277.61594926,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.51113369463245,
+    "estimated_full_shards_per_hour": 75.12709623006636,
     "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": [
-    {
-      "seconds": 1.0610393269998895,
-      "bytes": 5051210,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.4254050959998494,
       "bytes": 7441828,
@@ -562,6 +556,12 @@
       "bytes": 8430845,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.0249396090002847,
+      "bytes": 5941897,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -141,3 +141,4 @@
 {"step": 618500, "loss": 0.5130348801612854, "lr": 0.0001, "grad_norm": 0.13897861938266662, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90530910109, "tokens_per_second": 118051.8214918686, "data_wait_s": 1.4335000741994008e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 618600, "loss": 0.5336444973945618, "lr": 0.0001, "grad_norm": 0.1612524868048308, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90545772982, "tokens_per_second": 154279.9954747513, "data_wait_s": 1.6031000996008515e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 618700, "loss": 0.5049571990966797, "lr": 0.0001, "grad_norm": 0.16050327764789268, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90560455586, "tokens_per_second": 153288.01241546977, "data_wait_s": 1.1805999747593887e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 618800, "loss": 0.5379113554954529, "lr": 0.0001, "grad_norm": 0.15848362632127905, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90575060694, "tokens_per_second": 152261.77539880358, "data_wait_s": 1.1719001122401096e-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:c4456a5ff3c6df99b9965f0a048d73f60fc0d24eaacbe0970a207376651488ce
-size 305388
+oid sha256:4ee59f62a8e17c92e93e03ae5df3a543de493f47696070c3fc30b3d1d9767675
+size 307506