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": 6402.957566841184,
-      "required_pages_per_second_upper_estimate": 0.0005549921337462057
+      "required_token_positions_per_second": 71630.76008151886,
+      "required_pages_per_second_upper_estimate": 0.006208772737364407
     },
     {
       "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": 800.369695855148,
-      "required_pages_per_second_upper_estimate": 0.0003192794382699649
+      "required_token_positions_per_second": 8953.845010189858,
+      "required_pages_per_second_upper_estimate": 0.0035718226464775244
     },
     {
       "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": 1600.739391710296,
-      "required_pages_per_second_upper_estimate": 0.00037030288622107795
+      "required_token_positions_per_second": 17907.690020379716,
+      "required_pages_per_second_upper_estimate": 0.004142628921634695
     },
     {
       "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": 2401.109087565444,
-      "required_pages_per_second_upper_estimate": 0.00014621494578695045
+      "required_token_positions_per_second": 26861.535030569572,
+      "required_pages_per_second_upper_estimate": 0.001635726551779148
     },
     {
       "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": 2401.109087565444,
-      "required_pages_per_second_upper_estimate": 0.0001848440242561497
+      "required_token_positions_per_second": 26861.535030569572,
+      "required_pages_per_second_upper_estimate": 0.0020678753241413013
     },
     {
       "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": 3,
-      "required_token_positions_per_second": 2401.109087565444,
-      "required_pages_per_second_upper_estimate": 0.02981373887238715
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 26861.535030569572,
+      "required_pages_per_second_upper_estimate": 0.3335303653050098
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,20 +155,20 @@
   "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": 6,
+    "prefetch_batches": 59,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.04660634606349078,
-    "producer_p95_seconds": 0.10834533500002408,
-    "gpu_update_seconds": 11.514678838699965,
-    "producer_batches_per_second": 21.456305513367695,
-    "gpu_batches_per_second": 0.08684567012317145,
-    "producer_headroom": 247.06246705145637,
+    "producer_batch_seconds": 0.04685044286666847,
+    "producer_p95_seconds": 0.1082292760000314,
+    "gpu_update_seconds": 1.029278481983081,
+    "producer_batches_per_second": 21.344515415700485,
+    "gpu_batches_per_second": 0.9715543630848369,
+    "producer_headroom": 21.969450425736664,
     "sustainable": true,
-    "buffer_seconds": 69.08807303219979,
+    "buffer_seconds": 60.72743043700177,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 17,
+    "upload_pages_per_commit": 24,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -1 +1,2 @@
 {"step": 195100, "loss": 0.5759704150259495, "lr": 0.0003, "grad_norm": 0.14326037266666766, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28555098636, "tokens_per_second": 47929.59643830316, "data_wait_s": 1.7429000081392587e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 195200, "loss": 0.6118105687201023, "lr": 0.0003, "grad_norm": 0.13807380891249546, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28569686874, "tokens_per_second": 148719.4751744541, "data_wait_s": 1.639900006011885e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:512921ae630081cb0647480c9157eac9521b9916a28b8874aa22a3d13634eee4
-size 2283
+oid sha256:d66b75ab7cd70570cf280c16e8dd2952c33bcb85ee39091693a97547f099a4fc
+size 4401