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": 75025.43238111252,
-      "required_pages_per_second_upper_estimate": 0.005190433495112444
+      "required_token_positions_per_second": 75771.64240425773,
+      "required_pages_per_second_upper_estimate": 0.005242058035959429
     },
     {
       "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": 9378.179047639065,
-      "required_pages_per_second_upper_estimate": 0.003926447728598407
+      "required_token_positions_per_second": 9471.455300532216,
+      "required_pages_per_second_upper_estimate": 0.003965500547855113
     },
     {
       "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": 18756.35809527813,
-      "required_pages_per_second_upper_estimate": 0.0039064547528204685
+      "required_token_positions_per_second": 18942.910601064432,
+      "required_pages_per_second_upper_estimate": 0.003945308720055341
     },
     {
       "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": 28134.537142917194,
-      "required_pages_per_second_upper_estimate": 0.0016866225427670724
+      "required_token_positions_per_second": 28414.36590159665,
+      "required_pages_per_second_upper_estimate": 0.0017033978495761314
     },
     {
       "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": 28134.537142917194,
-      "required_pages_per_second_upper_estimate": 0.002187336512111303
+      "required_token_positions_per_second": 28414.36590159665,
+      "required_pages_per_second_upper_estimate": 0.002209091967262
     },
     {
       "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": 28134.537142917194,
-      "required_pages_per_second_upper_estimate": 0.3608796339569425
+      "required_token_positions_per_second": 28414.36590159665,
+      "required_pages_per_second_upper_estimate": 0.3644689768165705
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1360806367811449,
-    "producer_p95_seconds": 0.28600772700156085,
-    "gpu_update_seconds": 0.9827067656935071,
-    "producer_batches_per_second": 7.34858407231203,
-    "gpu_batches_per_second": 1.017597552912225,
-    "producer_headroom": 7.221503286128576,
+    "producer_batch_seconds": 0.09873860441408056,
+    "producer_p95_seconds": 0.2845255479987827,
+    "gpu_update_seconds": 0.9730289282452864,
+    "producer_batches_per_second": 10.127751004118867,
+    "gpu_batches_per_second": 1.027718674102888,
+    "producer_headroom": 9.854594705072905,
     "sustainable": true,
-    "buffer_seconds": 60.92781947299744,
+    "buffer_seconds": 60.327793551207755,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 344030.89137626847,
-    "upload_bytes_per_second": 5369334.261575148,
+    "generation_bytes_per_second": 357370.55697371974,
+    "upload_bytes_per_second": 5358293.928287545,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.82102066007653,
+    "estimated_full_shards_per_hour": 76.68340236576742,
     "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": 1.216705439001089,
-      "bytes": 6221386,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.2352482690002944,
       "bytes": 2570616,
@@ -562,6 +556,12 @@
       "bytes": 2755782,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.3326452839974081,
+      "bytes": 5962377,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -257,3 +257,4 @@
 {"step": 745700, "loss": 0.5213366746902466, "lr": 5e-05, "grad_norm": 0.22292530404676922, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109150134328, "tokens_per_second": 151497.57275360494, "data_wait_s": 1.5575998986605555e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 745800, "loss": 0.569585382938385, "lr": 5e-05, "grad_norm": 0.22521326874446224, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109164898701, "tokens_per_second": 151600.38380344637, "data_wait_s": 1.0897001629928127e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 745900, "loss": 0.5409356951713562, "lr": 5e-05, "grad_norm": 0.22679761135973822, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109179574178, "tokens_per_second": 151000.43837524575, "data_wait_s": 1.3753000530414283e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 746000, "loss": 0.5546980500221252, "lr": 5e-05, "grad_norm": 0.2263805816080096, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109194315552, "tokens_per_second": 151491.65458015518, "data_wait_s": 1.2363001587800682e-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:cbcb9fc22e23ef12a55a650482d9ffb1335702d1bb8befb0371e8ddab8c4df7d
-size 552642
+oid sha256:b67124e799f3b9ffadd2f4183a3c44c98e21ed15bcb64a48b89a8c268a4e24e1
+size 554760