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": 76671.86685246631,
-      "required_pages_per_second_upper_estimate": 0.0053043376520950005
+      "required_token_positions_per_second": 76116.24921334932,
+      "required_pages_per_second_upper_estimate": 0.005265898734610346
     },
     {
       "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": 9583.983356558289,
-      "required_pages_per_second_upper_estimate": 0.00401261369506021
+      "required_token_positions_per_second": 9514.531151668665,
+      "required_pages_per_second_upper_estimate": 0.003983535507199884
     },
     {
       "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": 19167.966713116577,
-      "required_pages_per_second_upper_estimate": 0.003992181973067008
+      "required_token_positions_per_second": 19029.06230333733,
+      "required_pages_per_second_upper_estimate": 0.003963251847665613
     },
     {
       "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": 28751.950069674866,
-      "required_pages_per_second_upper_estimate": 0.0017236355049912418
+      "required_token_positions_per_second": 28543.593455005997,
+      "required_pages_per_second_upper_estimate": 0.0017111448440839728
     },
     {
       "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": 28751.950069674866,
-      "required_pages_per_second_upper_estimate": 0.002235337651454252
+      "required_token_positions_per_second": 28543.593455005997,
+      "required_pages_per_second_upper_estimate": 0.0022191388411276405
     },
     {
       "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": 28751.950069674866,
-      "required_pages_per_second_upper_estimate": 0.3687991440550386
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28543.593455005997,
+      "required_pages_per_second_upper_estimate": 0.3661265691179692
     }
   ],
   "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": 0.11959173422724234,
-    "producer_p95_seconds": 0.30257836999953724,
-    "gpu_update_seconds": 0.9616043410273163,
-    "producer_batches_per_second": 8.361781911280332,
-    "gpu_batches_per_second": 1.0399287496265504,
-    "producer_headroom": 8.040725784610856,
+    "producer_batch_seconds": 0.4402515033517602,
+    "producer_p95_seconds": 0.2831134790030774,
+    "gpu_update_seconds": 0.9686236613334006,
+    "producer_batches_per_second": 2.271428927298862,
+    "gpu_batches_per_second": 1.032392703089048,
+    "producer_headroom": 2.2001598040188224,
     "sustainable": true,
-    "buffer_seconds": 60.581073484720925,
+    "buffer_seconds": 60.05466700267084,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 439101.36498152063,
+    "generation_bytes_per_second": 390368.8037081872,
     "upload_bytes_per_second": 4798834.12788914,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.22093116840567,
+    "estimated_full_shards_per_hour": 83.76405795511448,
     "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-v1/1791143623306220418-eb05a537/train.jsonl

@@ -308,3 +308,4 @@
 {"step": 715800, "loss": 0.5447097420692444, "lr": 8e-05, "grad_norm": 0.219595896913659, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104773364590, "tokens_per_second": 150229.57645654277, "data_wait_s": 1.3037999451626092e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 715900, "loss": 0.507247269153595, "lr": 8e-05, "grad_norm": 0.2181759101612972, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104787960016, "tokens_per_second": 151484.94802136908, "data_wait_s": 1.7903999832924455e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 716000, "loss": 0.5255452990531921, "lr": 8e-05, "grad_norm": 0.2099402970839708, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104802603951, "tokens_per_second": 152256.716971786, "data_wait_s": 1.0746000043582171e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 716100, "loss": 0.542158305644989, "lr": 8e-05, "grad_norm": 0.23822071220199037, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104817278984, "tokens_per_second": 152792.09080238626, "data_wait_s": 1.608499587746337e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3c57279280c11086af9ed006cdc84172a7a218f373fbfb6269e652687fc0889e
-size 661443
+oid sha256:b9f09becc11262ac377428a41238587aae03435047ee146b639f7a66bd723460
+size 663561