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": 76839.30006210206,
-      "required_pages_per_second_upper_estimate": 0.0053159210700361725
+      "required_token_positions_per_second": 75712.31936436366,
+      "required_pages_per_second_upper_estimate": 0.005237953930411141
     },
     {
       "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": 9604.912507762758,
-      "required_pages_per_second_upper_estimate": 0.004021376293619145
+      "required_token_positions_per_second": 9464.039920545458,
+      "required_pages_per_second_upper_estimate": 0.003962395883105401
     },
     {
       "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": 19209.825015525515,
-      "required_pages_per_second_upper_estimate": 0.0040008999535811225
+      "required_token_positions_per_second": 18928.079841090916,
+      "required_pages_per_second_upper_estimate": 0.003942219863866299
     },
     {
       "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": 28814.737523288273,
-      "required_pages_per_second_upper_estimate": 0.001727399516964475
+      "required_token_positions_per_second": 28392.11976163637,
+      "required_pages_per_second_upper_estimate": 0.0017020642274534015
     },
     {
       "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": 28814.737523288273,
-      "required_pages_per_second_upper_estimate": 0.002240219099799873
+      "required_token_positions_per_second": 28392.11976163637,
+      "required_pages_per_second_upper_estimate": 0.0022073624277305164
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28814.737523288273,
-      "required_pages_per_second_upper_estimate": 0.3696045140940762
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28392.11976163637,
+      "required_pages_per_second_upper_estimate": 0.3641836272192041
     }
   ],
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10693364155460472,
-    "producer_p95_seconds": 0.2741348730014579,
-    "gpu_update_seconds": 0.959509000477783,
-    "producer_batches_per_second": 9.351593992890992,
-    "gpu_batches_per_second": 1.0421997078735632,
-    "producer_headroom": 8.972938604992875,
+    "producer_batch_seconds": 0.13644720317179804,
+    "producer_p95_seconds": 0.2923337749998609,
+    "gpu_update_seconds": 0.9737913277386977,
+    "producer_batches_per_second": 7.328842048458255,
+    "gpu_batches_per_second": 1.0269140538786303,
+    "producer_headroom": 7.136762829155361,
     "sustainable": true,
-    "buffer_seconds": 60.44906703010033,
+    "buffer_seconds": 60.375062319799255,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 288681.11789956,
+    "generation_bytes_per_second": 295226.75045665295,
     "upload_bytes_per_second": 5012878.109568305,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.94424774875737,
+    "estimated_full_shards_per_hour": 63.348788120982086,
     "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-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -260,3 +260,4 @@
 {"step": 951000, "loss": 0.5581480264663696, "lr": 3e-05, "grad_norm": 0.34757897631609186, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139200565068, "tokens_per_second": 153829.92213001658, "data_wait_s": 1.072399754775688e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 951100, "loss": 0.5073821544647217, "lr": 3e-05, "grad_norm": 0.3306391893619082, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139215124138, "tokens_per_second": 119902.04435433538, "data_wait_s": 7.883998478064314e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 951200, "loss": 0.5129421353340149, "lr": 3e-05, "grad_norm": 0.33315655108107156, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139229666414, "tokens_per_second": 150673.39371447515, "data_wait_s": 1.5237001207424328e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 951300, "loss": 0.5131524205207825, "lr": 3e-05, "grad_norm": 0.34254456535394895, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139244193554, "tokens_per_second": 151113.32410119823, "data_wait_s": 1.3173998013371602e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1a7afab951fe167f19fccfcbcece2d4d8c9aed94e3c7ba97064465c2242eb873
-size 558996
+oid sha256:3b630a358288184f503d70bc5f1dcc0f542de43f5f8231d42836d75f81640ace
+size 561114