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": 76109.09255933682,
-      "required_pages_per_second_upper_estimate": 0.006596943246311771
+      "required_token_positions_per_second": 74255.66936111983,
+      "required_pages_per_second_upper_estimate": 0.006436293220948472
     },
     {
       "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": 9513.636569917102,
-      "required_pages_per_second_upper_estimate": 0.003795131869282393
+      "required_token_positions_per_second": 9281.958670139978,
+      "required_pages_per_second_upper_estimate": 0.0037027120911680145
     },
     {
       "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": 19027.273139834204,
-      "required_pages_per_second_upper_estimate": 0.004401624772330564
+      "required_token_positions_per_second": 18563.917340279957,
+      "required_pages_per_second_upper_estimate": 0.00429443556288724
     },
     {
       "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": 28540.909709751304,
-      "required_pages_per_second_upper_estimate": 0.0017379916587433237
+      "required_token_positions_per_second": 27845.876010419935,
+      "required_pages_per_second_upper_estimate": 0.0016956677530140427
     },
     {
       "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": 28540.909709751304,
-      "required_pages_per_second_upper_estimate": 0.0021971582357513597
+      "required_token_positions_per_second": 27845.876010419935,
+      "required_pages_per_second_upper_estimate": 0.002143652617600411
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28540.909709751304,
-      "required_pages_per_second_upper_estimate": 0.3543825783149522
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27845.876010419935,
+      "required_pages_per_second_upper_estimate": 0.3457525858974128
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.052114650429693654,
-    "producer_p95_seconds": 0.19034486900000047,
-    "gpu_update_seconds": 0.9687147424930806,
-    "producer_batches_per_second": 19.188462203139416,
-    "gpu_batches_per_second": 1.0322956347566299,
-    "producer_headroom": 18.58814622195241,
+    "producer_batch_seconds": 0.251848909101561,
+    "producer_p95_seconds": 0.19614139399982378,
+    "gpu_update_seconds": 0.992893884525454,
+    "producer_batches_per_second": 3.970634629974666,
+    "gpu_batches_per_second": 1.007156973756508,
+    "producer_headroom": 3.9424188417868353,
     "sustainable": true,
-    "buffer_seconds": 60.060314034571,
+    "buffer_seconds": 60.5665269560527,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 272416.2244608926,
+    "generation_bytes_per_second": 260509.09576488013,
     "upload_bytes_per_second": 4451938.596442435,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.4541802441605,
+    "estimated_full_shards_per_hour": 55.89918760976603,
     "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-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -58,3 +58,4 @@
 {"step": 5800, "loss": 0.7632534131407738, "lr": 0.0003, "grad_norm": 0.16977647062289472, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 848092174, "tokens_per_second": 147800.9207885724, "data_wait_s": 1.1238999832130503e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 5900, "loss": 0.7534791063517332, "lr": 0.0003, "grad_norm": 0.15907890939786187, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 862772198, "tokens_per_second": 149378.6033500232, "data_wait_s": 1.3300000318849925e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 6000, "loss": 0.7702555432915688, "lr": 0.0003, "grad_norm": 0.17842406945478048, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 877349814, "tokens_per_second": 149471.56779400317, "data_wait_s": 1.1080999684054404e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 6100, "loss": 0.7579893618822098, "lr": 0.0003, "grad_norm": 0.16477188436430226, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 891895665, "tokens_per_second": 139470.8196672479, "data_wait_s": 1.0054999620479066e-05, "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 sha256:7dcb8c6d5606e7f3172ea30ec5646cf6358813c0169143b98c05385acd891309
-size 125879
+oid sha256:5b08c9a7d4b36d73785f5f2f0035b3986415fe8f807ccbb9adcd1cc9bd6f140b
+size 127962