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": 77389.27735885547,
-      "required_pages_per_second_upper_estimate": 0.005353969775548717
+      "required_token_positions_per_second": 77596.1092437681,
+      "required_pages_per_second_upper_estimate": 0.005368278885263067
     },
     {
       "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": 9673.659669856934,
-      "required_pages_per_second_upper_estimate": 0.004050159294783984
+      "required_token_positions_per_second": 9699.513655471013,
+      "required_pages_per_second_upper_estimate": 0.0040609838186679865
     },
     {
       "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": 19347.319339713868,
-      "required_pages_per_second_upper_estimate": 0.004029536395340392
+      "required_token_positions_per_second": 19399.027310942027,
+      "required_pages_per_second_upper_estimate": 0.004040305802116315
     },
     {
       "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": 29020.9790095708,
-      "required_pages_per_second_upper_estimate": 0.0017397633791545971
+      "required_token_positions_per_second": 29098.540966413042,
+      "required_pages_per_second_upper_estimate": 0.0017444131000370388
     },
     {
       "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": 29020.9790095708,
-      "required_pages_per_second_upper_estimate": 0.002256253468197913
+      "required_token_positions_per_second": 29098.540966413042,
+      "required_pages_per_second_upper_estimate": 0.002262283569183404
     },
     {
       "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": 29020.9790095708,
-      "required_pages_per_second_upper_estimate": 0.37224995843525355
+      "required_token_positions_per_second": 29098.540966413042,
+      "required_pages_per_second_upper_estimate": 0.373244839938085
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 720750,
-    "producer_batch_seconds": 0.4578429652029854,
-    "producer_p95_seconds": 0.2389157139987219,
-    "gpu_update_seconds": 0.9526901208564326,
-    "producer_batches_per_second": 2.1841549963679108,
-    "gpu_batches_per_second": 1.049659252371629,
-    "producer_headroom": 2.080822887458926,
+    "producer_batch_seconds": 0.06824385599986726,
+    "producer_p95_seconds": 0.23904048499753117,
+    "gpu_update_seconds": 0.9501507320216733,
+    "producer_batches_per_second": 14.653333774134115,
+    "gpu_batches_per_second": 1.0524645893523235,
+    "producer_headroom": 13.922875812051437,
     "sustainable": true,
-    "buffer_seconds": 60.01947761395525,
+    "buffer_seconds": 60.80964684938709,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 299641.3044262727,
-    "upload_bytes_per_second": 5592687.288932413,
+    "generation_bytes_per_second": 300012.90759855305,
+    "upload_bytes_per_second": 5632667.379678929,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.2960486373056,
+    "estimated_full_shards_per_hour": 64.37578602759784,
     "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": [
-    {
-      "seconds": 1.3078732329995546,
-      "bytes": 7188440,
-      "pages": 42,
-      "shards": 2
-    },
     {
       "seconds": 1.37187312900096,
       "bytes": 11914910,
@@ -562,6 +556,12 @@
       "bytes": 2738639,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.369231013002718,
+      "bytes": 10621268,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -225,3 +225,4 @@
 {"step": 707500, "loss": 0.5464622974395752, "lr": 8e-05, "grad_norm": 0.19706057669124938, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103557788498, "tokens_per_second": 152070.82317731014, "data_wait_s": 1.650600097491406e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 707600, "loss": 0.5644626617431641, "lr": 8e-05, "grad_norm": 0.23308038512785945, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103572462469, "tokens_per_second": 134158.50991644702, "data_wait_s": 1.487299959990196e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 707700, "loss": 0.5445774793624878, "lr": 8e-05, "grad_norm": 0.23317238347965258, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103587149186, "tokens_per_second": 153324.85007495372, "data_wait_s": 1.4682002074550837e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 707800, "loss": 0.5774579644203186, "lr": 8e-05, "grad_norm": 0.20599089814819257, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103601807208, "tokens_per_second": 152264.95125941405, "data_wait_s": 1.475400131312199e-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 sha256:09d4698f0f5599956a6a742838fcdbdd90ab2eb02515a915ed356161b260d6f8
-size 484083
+oid sha256:06f9869529bbb7024ee710903b22ec3e02f561cec467567b8f2f45586d17a18e
+size 486201