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": 76323.04476482134,
-      "required_pages_per_second_upper_estimate": 0.005280205330692967
+      "required_token_positions_per_second": 76749.23718949944,
+      "required_pages_per_second_upper_estimate": 0.005309690311534868
     },
     {
       "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": 9540.380595602668,
-      "required_pages_per_second_upper_estimate": 0.00399435812957728
+      "required_token_positions_per_second": 9593.65464868743,
+      "required_pages_per_second_upper_estimate": 0.0040166628631151365
     },
     {
       "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": 19080.761191205336,
-      "required_pages_per_second_upper_estimate": 0.003974019362617162
+      "required_token_positions_per_second": 19187.30929737486,
+      "required_pages_per_second_upper_estimate": 0.003996210523269758
     },
     {
       "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": 28621.141786808003,
-      "required_pages_per_second_upper_estimate": 0.0017157937481660564
+      "required_token_positions_per_second": 28780.963946062293,
+      "required_pages_per_second_upper_estimate": 0.001725374842579044
     },
     {
       "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": 28621.141786808003,
-      "required_pages_per_second_upper_estimate": 0.0022251678828261834
+      "required_token_positions_per_second": 28780.963946062293,
+      "required_pages_per_second_upper_estimate": 0.0022375933527248194
     },
     {
       "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": 28621.141786808003,
-      "required_pages_per_second_upper_estimate": 0.3671212758534139
+      "required_token_positions_per_second": 28780.963946062293,
+      "required_pages_per_second_upper_estimate": 0.36917130290866323
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08014884166411207,
-    "producer_p95_seconds": 0.2734274979993643,
-    "gpu_update_seconds": 0.9659991975841948,
-    "producer_batches_per_second": 12.47678667884936,
-    "gpu_batches_per_second": 1.0351975472659145,
-    "producer_headroom": 12.052565920197653,
+    "producer_batch_seconds": 0.43636395428126207,
+    "producer_p95_seconds": 0.30716543999733403,
+    "gpu_update_seconds": 0.9606349548199444,
+    "producer_batches_per_second": 2.2916649970484078,
+    "gpu_batches_per_second": 1.0409781519843133,
+    "producer_headroom": 2.2014535009020455,
     "sustainable": true,
-    "buffer_seconds": 60.85794944780427,
+    "buffer_seconds": 60.5200021536565,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 431906.85572984786,
+    "generation_bytes_per_second": 380184.4270397572,
     "upload_bytes_per_second": 5394312.85270678,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.67715696260049,
+    "estimated_full_shards_per_hour": 81.5787278022245,
     "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-v1/1791039222312944374-63c10d34/train.jsonl

@@ -292,3 +292,4 @@
 {"step": 633600, "loss": 0.5683121085166931, "lr": 0.0001, "grad_norm": 0.16822814200641525, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92742428794, "tokens_per_second": 151771.97313894506, "data_wait_s": 1.267100014956668e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 633700, "loss": 0.6104334592819214, "lr": 0.0001, "grad_norm": 0.17951660589050078, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92757019066, "tokens_per_second": 150911.3407158322, "data_wait_s": 1.27929997688625e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 633800, "loss": 0.5685456395149231, "lr": 0.0001, "grad_norm": 0.18621276709977438, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92771598291, "tokens_per_second": 151822.60911483734, "data_wait_s": 1.2259999493835494e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 633900, "loss": 0.5527037382125854, "lr": 0.0001, "grad_norm": 0.17776146793990177, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92786218479, "tokens_per_second": 151091.99041596107, "data_wait_s": 1.7308997485088184e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:92f6bb534947fa0653e6facf57eef9edc7b8d5f504ae9e88754808f630333b10
-size 627555
+oid sha256:a3427e1dce3450afa2c84f3a032ca0f3c0534d0c6d88ec26701d01f5d57a19fd
+size 629673