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": 76342.26255933823,
-      "required_pages_per_second_upper_estimate": 0.005281534862309072
+      "required_token_positions_per_second": 76033.1486420992,
+      "required_pages_per_second_upper_estimate": 0.005260149644271372
     },
     {
       "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": 9542.782819917278,
-      "required_pages_per_second_upper_estimate": 0.003995363890733659
+      "required_token_positions_per_second": 9504.1435802624,
+      "required_pages_per_second_upper_estimate": 0.003979186447969239
     },
     {
       "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": 19085.565639834556,
-      "required_pages_per_second_upper_estimate": 0.0039750200025647974
+      "required_token_positions_per_second": 19008.2871605248,
+      "required_pages_per_second_upper_estimate": 0.003958924933295122
     },
     {
       "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": 28628.348459751833,
-      "required_pages_per_second_upper_estimate": 0.0017162257772050866
+      "required_token_positions_per_second": 28512.430740787204,
+      "required_pages_per_second_upper_estimate": 0.0017092766869492655
     },
     {
       "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": 28628.348459751833,
-      "required_pages_per_second_upper_estimate": 0.002225728169948762
+      "required_token_positions_per_second": 28512.430740787204,
+      "required_pages_per_second_upper_estimate": 0.002216716077167423
     },
     {
       "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": 28628.348459751833,
-      "required_pages_per_second_upper_estimate": 0.3672137153160148
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28512.430740787204,
+      "required_pages_per_second_upper_estimate": 0.36572684727988614
     }
   ],
   "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": 0.09203488360151368,
-    "producer_p95_seconds": 0.2504416920019139,
-    "gpu_update_seconds": 0.9657560246225838,
-    "producer_batches_per_second": 10.865445371015314,
-    "gpu_batches_per_second": 1.035458205286163,
-    "producer_headroom": 10.493369327265604,
+    "producer_batch_seconds": 0.13689437425776418,
+    "producer_p95_seconds": 0.27760965599736664,
+    "gpu_update_seconds": 0.9696823203659508,
+    "producer_batches_per_second": 7.304902085436016,
+    "gpu_batches_per_second": 1.0312655794555556,
+    "producer_headroom": 7.083434404251669,
     "sustainable": true,
-    "buffer_seconds": 60.84262955122278,
+    "buffer_seconds": 60.12030386268895,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 514709.2665774475,
+    "generation_bytes_per_second": 515958.6055254786,
     "upload_bytes_per_second": 3936235.4823358366,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 110.44462678902214,
+    "estimated_full_shards_per_hour": 110.71270584414738,
     "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-v1/1791300814123050567-72be911d/train.jsonl

@@ -230,3 +230,4 @@
 {"step": 823000, "loss": 0.5252871513366699, "lr": 3e-05, "grad_norm": 0.26089176789618596, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120463834068, "tokens_per_second": 150460.31595709507, "data_wait_s": 1.3984001270728186e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 823100, "loss": 0.5351876616477966, "lr": 3e-05, "grad_norm": 0.2871865533724164, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120478645494, "tokens_per_second": 152926.0526024939, "data_wait_s": 1.365699790767394e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 823200, "loss": 0.5542900562286377, "lr": 3e-05, "grad_norm": 0.2848713946231784, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120493358351, "tokens_per_second": 152233.3260610851, "data_wait_s": 9.142000635620207e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 823300, "loss": 0.5318233370780945, "lr": 3e-05, "grad_norm": 0.2707454232859008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120508057249, "tokens_per_second": 152149.08682513738, "data_wait_s": 2.482100171619095e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:796c850d075d7093268a704ea38f298be3255743de4923196e1d9a7edb9ce313
-size 494673
+oid sha256:8f5f8aff1a9a62b0cf2e931aa8ffe50af01c1cae03e70c2aa1b8aa557741a07a
+size 496791