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": 75522.53967562613,
-      "required_pages_per_second_upper_estimate": 0.006546102328956422
+      "required_token_positions_per_second": 76202.75147966256,
+      "required_pages_per_second_upper_estimate": 0.0066050613641492435
     },
     {
       "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": 9440.317459453267,
-      "required_pages_per_second_upper_estimate": 0.0037658837798999786
+      "required_token_positions_per_second": 9525.34393495782,
+      "required_pages_per_second_upper_estimate": 0.003799802112237841
     },
     {
       "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": 18880.634918906533,
-      "required_pages_per_second_upper_estimate": 0.004367702600663492
+      "required_token_positions_per_second": 19050.68786991564,
+      "required_pages_per_second_upper_estimate": 0.004407041358049729
     },
     {
       "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": 28320.9523783598,
-      "required_pages_per_second_upper_estimate": 0.001724597411288511
+      "required_token_positions_per_second": 28576.031804873463,
+      "required_pages_per_second_upper_estimate": 0.0017401304100648713
     },
     {
       "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": 28320.9523783598,
-      "required_pages_per_second_upper_estimate": 0.0021802253114999786
+      "required_token_positions_per_second": 28576.031804873463,
+      "required_pages_per_second_upper_estimate": 0.0021998620318580456
     },
     {
       "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": 28320.9523783598,
-      "required_pages_per_second_upper_estimate": 0.3516514444089028
+      "required_token_positions_per_second": 28576.031804873463,
+      "required_pages_per_second_upper_estimate": 0.35481867719027854
     }
   ],
   "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": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08573827935944678,
-    "producer_p95_seconds": 0.2019061650007643,
-    "gpu_update_seconds": 0.9762383563458831,
-    "producer_batches_per_second": 11.663401778890707,
-    "gpu_batches_per_second": 1.0243400021108147,
-    "producer_headroom": 11.386260182025913,
+    "producer_batch_seconds": 0.06063768392968427,
+    "producer_p95_seconds": 0.19892775200059987,
+    "gpu_update_seconds": 0.9675241191215642,
+    "producer_batches_per_second": 16.49139504008109,
+    "gpu_batches_per_second": 1.033565965164694,
+    "producer_headroom": 15.95582245924019,
     "sustainable": true,
-    "buffer_seconds": 60.52677809344475,
+    "buffer_seconds": 60.95401950465855,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 534748.2043081002,
+    "generation_bytes_per_second": 520797.13506103976,
     "upload_bytes_per_second": 5293928.85370719,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 114.74451634342435,
+    "estimated_full_shards_per_hour": 111.75094164727588,
     "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-v1/1790563279934888633-82973089/train.jsonl

@@ -135,3 +135,4 @@
 {"step": 168500, "loss": 0.5927061382681131, "lr": 0.0003, "grad_norm": 0.1414547014403755, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24660830463, "tokens_per_second": 150858.18277241423, "data_wait_s": 1.2579999747686088e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 168600, "loss": 0.6325449906289577, "lr": 0.0003, "grad_norm": 0.14246491317830984, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24675494207, "tokens_per_second": 149596.60712256024, "data_wait_s": 1.1685999197652563e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 168700, "loss": 0.6042628977447748, "lr": 0.0003, "grad_norm": 0.15650584227774014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24690141416, "tokens_per_second": 150191.7518256853, "data_wait_s": 1.3741000657319091e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 168800, "loss": 0.6158970892429352, "lr": 0.0003, "grad_norm": 0.18008997262078796, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24704702692, "tokens_per_second": 149343.14134793746, "data_wait_s": 1.2721000530291349e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f7b504728867ac45cf6edc836524b6bc81e032159e1787e2184bf9ecd18b0ac7
-size 291897
+oid sha256:43b32e665bb79cb27abdf2ca923b81755db7dc43767f2a19936f608216beb918
+size 294015