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": 77159.15043855768,
-      "required_pages_per_second_upper_estimate": 0.005338049061234495
+      "required_token_positions_per_second": 77563.09033818691,
+      "required_pages_per_second_upper_estimate": 0.005365994560760543
     },
     {
       "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": 9644.89380481971,
-      "required_pages_per_second_upper_estimate": 0.004038115627792468
+      "required_token_positions_per_second": 9695.386292273364,
+      "required_pages_per_second_upper_estimate": 0.004059255777886275
     },
     {
       "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": 19289.78760963942,
-      "required_pages_per_second_upper_estimate": 0.004017554053179635
+      "required_token_positions_per_second": 19390.77258454673,
+      "required_pages_per_second_upper_estimate": 0.004038586560299949
     },
     {
       "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": 28934.681414459134,
-      "required_pages_per_second_upper_estimate": 0.0017345899700964508
+      "required_token_positions_per_second": 29086.15887682009,
+      "required_pages_per_second_upper_estimate": 0.0017436708126722972
     },
     {
       "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": 28934.681414459134,
-      "required_pages_per_second_upper_estimate": 0.0022495442097609864
+      "required_token_positions_per_second": 29086.15887682009,
+      "required_pages_per_second_upper_estimate": 0.002261320916180608
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28934.681414459134,
-      "required_pages_per_second_upper_estimate": 0.3711430255442995
+      "required_token_positions_per_second": 29086.15887682009,
+      "required_pages_per_second_upper_estimate": 0.37308601578763856
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.10151906509380382,
-    "producer_p95_seconds": 0.18504816699987714,
-    "gpu_update_seconds": 0.9555315161059227,
-    "producer_batches_per_second": 9.85036652057422,
-    "gpu_batches_per_second": 1.0465379562521389,
-    "producer_headroom": 9.412335655603306,
+    "producer_batch_seconds": 0.049100155007820945,
+    "producer_p95_seconds": 0.162461720000465,
+    "gpu_update_seconds": 0.9505552148391029,
+    "producer_batches_per_second": 20.366534481219347,
+    "gpu_batches_per_second": 1.052016741783134,
+    "producer_headroom": 19.35951555932345,
     "sustainable": true,
-    "buffer_seconds": 60.19848551467313,
+    "buffer_seconds": 60.835533749702584,
     "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": 363818.30165771,
-    "upload_bytes_per_second": 4245835.198740466,
+    "generation_bytes_per_second": 385224.3194908117,
+    "upload_bytes_per_second": 4245126.185163211,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.06693827913737,
+    "estimated_full_shards_per_hour": 82.6601713995291,
     "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": [
@@ -484,6 +484,12 @@
       "bytes": 1133265,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.8210042499995325,
+      "bytes": 7672821,
+      "pages": 32,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -65,3 +65,4 @@
 {"step": 446700, "loss": 0.539176344871521, "lr": 0.0003, "grad_norm": 0.15264892739942929, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65383176323, "tokens_per_second": 152023.39627403402, "data_wait_s": 1.4923999515303876e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 446800, "loss": 0.5781931281089783, "lr": 0.0003, "grad_norm": 0.14260654019291844, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65397905096, "tokens_per_second": 152879.43145148747, "data_wait_s": 1.5163999705691822e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 446900, "loss": 0.6321390271186829, "lr": 0.0003, "grad_norm": 0.15205760743964034, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65412524072, "tokens_per_second": 152354.3841680319, "data_wait_s": 1.3597000361187384e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 447000, "loss": 0.4998493194580078, "lr": 0.0003, "grad_norm": 0.1414208802714209, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65427210081, "tokens_per_second": 152827.77021420054, "data_wait_s": 1.2985000466869678e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1ec3ac3b95f5c328b5f4afcede72fb7a0c339eaafa079c550cb4da179d583a6c
-size 142854
+oid sha256:ec3089ae545222f136d99f2be1274f8dd7283929afaf964fac5c8447eb392ddb
+size 144972