Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75562.24394015495,
-      "required_pages_per_second_upper_estimate": 0.006549543793976241
+      "required_token_positions_per_second": 75692.27071928447,
+      "required_pages_per_second_upper_estimate": 0.006560814185641329
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9445.280492519369,
-      "required_pages_per_second_upper_estimate": 0.003767863607994004
+      "required_token_positions_per_second": 9461.533839910559,
+      "required_pages_per_second_upper_estimate": 0.0037743473112775484
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18890.560985038737,
-      "required_pages_per_second_upper_estimate": 0.0043699988213703805
+      "required_token_positions_per_second": 18923.067679821117,
+      "required_pages_per_second_upper_estimate": 0.004377518673109995
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28335.841477558108,
-      "required_pages_per_second_upper_estimate": 0.0017255040793126217
+      "required_token_positions_per_second": 28384.601519731677,
+      "required_pages_per_second_upper_estimate": 0.001728473310056822
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28335.841477558108,
-      "required_pages_per_second_upper_estimate": 0.0021813715155719286
+      "required_token_positions_per_second": 28384.601519731677,
+      "required_pages_per_second_upper_estimate": 0.0021851251985948858
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28335.841477558108,
-      "required_pages_per_second_upper_estimate": 0.3518363171903362
+      "required_token_positions_per_second": 28384.601519731677,
+      "required_pages_per_second_upper_estimate": 0.35244175372476844
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.1869404895084017,
-    "producer_p95_seconds": 0.42604548599774716,
-    "gpu_update_seconds": 0.9757253908233897,
-    "producer_batches_per_second": 5.349295931714445,
-    "gpu_batches_per_second": 1.024878525663994,
-    "producer_headroom": 5.219443863602045,
+    "producer_batch_seconds": 0.12610325951567347,
+    "producer_p95_seconds": 0.42072688299958827,
+    "gpu_update_seconds": 0.9740492562765193,
+    "producer_batches_per_second": 7.93000913569335,
+    "gpu_batches_per_second": 1.026642126726406,
+    "producer_headroom": 7.724219500888111,
     "sustainable": true,
-    "buffer_seconds": 60.494974231050165,
+    "buffer_seconds": 60.391053889144196,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 433029.63237572496,
-    "upload_bytes_per_second": 5078253.661915997,
+    "generation_bytes_per_second": 455207.49298486306,
+    "upload_bytes_per_second": 5054314.976417552,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.91807869390308,
+    "estimated_full_shards_per_hour": 97.67693130645198,
     "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": 1.0482589550010744,
-      "bytes": 7056720,
-      "pages": 42,
-      "shards": 2
-    },
     {
       "seconds": 1.9086563699966064,
       "bytes": 7678172,
@@ -562,6 +556,12 @@
       "bytes": 7585529,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.1198065829958068,
+      "bytes": 5440269,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -421,3 +421,4 @@
 {"step": 42100, "loss": 0.6266840100288391, "lr": 0.0003, "grad_norm": 0.1313392719859586, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6160514075, "tokens_per_second": 147259.96204561682, "data_wait_s": 1.247199543286115e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 42200, "loss": 0.6159834675490856, "lr": 0.0003, "grad_norm": 0.12582849847270944, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6175109717, "tokens_per_second": 147744.1308115539, "data_wait_s": 1.2521995813585818e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 42300, "loss": 0.637635163962841, "lr": 0.0003, "grad_norm": 0.12041313327317661, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6189763884, "tokens_per_second": 149208.78923071647, "data_wait_s": 1.2611999409273267e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 42400, "loss": 0.6545305550098419, "lr": 0.0003, "grad_norm": 0.13580467009364228, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6204361023, "tokens_per_second": 148229.16309410866, "data_wait_s": 9.611998393666e-06, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4c7402ed2c46e5f3d6028bfa8d42ed9dc3b576eec787ac40e95d83e0d57934a7
-size 896569
+oid sha256:a92e9c96bd6b1b273b2227c7b9a0a29ccc78c5f9f90fb829a09a24a2c0362ab4
+size 898687