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": 76663.52245458492,
-      "required_pages_per_second_upper_estimate": 0.005303760367287896
+      "required_token_positions_per_second": 76322.0994637896,
+      "required_pages_per_second_upper_estimate": 0.00528013993257419
     },
     {
       "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": 9582.940306823115,
-      "required_pages_per_second_upper_estimate": 0.004012176991917448
+      "required_token_positions_per_second": 9540.2624329737,
+      "required_pages_per_second_upper_estimate": 0.003994308657352047
     },
     {
       "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": 19165.88061364623,
-      "required_pages_per_second_upper_estimate": 0.003991747493561477
+      "required_token_positions_per_second": 19080.5248659474,
+      "required_pages_per_second_upper_estimate": 0.003973970142298249
     },
     {
       "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": 28748.820920469345,
-      "required_pages_per_second_upper_estimate": 0.001723447917274305
+      "required_token_positions_per_second": 28620.787298921095,
+      "required_pages_per_second_upper_estimate": 0.0017157724971585081
     },
     {
       "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": 28748.820920469345,
-      "required_pages_per_second_upper_estimate": 0.002235094373867196
+      "required_token_positions_per_second": 28620.787298921095,
+      "required_pages_per_second_upper_estimate": 0.0022251403229522036
     },
     {
       "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": 28748.820920469345,
-      "required_pages_per_second_upper_estimate": 0.36875900668884887
+      "required_token_positions_per_second": 28620.787298921095,
+      "required_pages_per_second_upper_estimate": 0.3671167288634201
     }
   ],
   "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": 0.2902776095155559,
-    "producer_p95_seconds": 0.2796941440028604,
-    "gpu_update_seconds": 0.9617090063096969,
-    "producer_batches_per_second": 3.444978073468702,
-    "gpu_batches_per_second": 1.039815571486883,
-    "producer_headroom": 3.3130664397942793,
+    "producer_batch_seconds": 0.407357053531058,
+    "producer_p95_seconds": 0.28313544599950546,
+    "gpu_update_seconds": 0.9660111621402614,
+    "producer_batches_per_second": 2.454848863746893,
+    "gpu_batches_per_second": 1.0351847258000975,
+    "producer_headroom": 2.371411403746836,
     "sustainable": true,
-    "buffer_seconds": 60.587667397510906,
+    "buffer_seconds": 60.85870321483647,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 402806.9416905692,
+    "generation_bytes_per_second": 359286.0486831134,
     "upload_bytes_per_second": 5341414.979204587,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.4329928211003,
+    "estimated_full_shards_per_hour": 77.09442229623843,
     "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

@@ -297,3 +297,4 @@
 {"step": 634100, "loss": 0.5661565065383911, "lr": 0.0001, "grad_norm": 0.16171603905457355, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92815519092, "tokens_per_second": 153121.16051451332, "data_wait_s": 1.121400055126287e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 634200, "loss": 0.6155643463134766, "lr": 0.0001, "grad_norm": 0.20585343553527108, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92830359763, "tokens_per_second": 154408.0504682232, "data_wait_s": 1.1296000593574718e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 634300, "loss": 0.5496290922164917, "lr": 0.0001, "grad_norm": 0.16686165868952338, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92845059621, "tokens_per_second": 152349.60941087644, "data_wait_s": 1.1431999155320227e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 634400, "loss": 0.542793869972229, "lr": 0.0001, "grad_norm": 0.17187412109977634, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92859536246, "tokens_per_second": 150589.42568573117, "data_wait_s": 1.032799991662614e-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:d295e8c18a532044ad1a5704581f0e9719edd27d66112d52f9b373973c7cbefc
-size 638145
+oid sha256:b9390719e97fbf3dc2246942af0caaf41f498d708ab3cd44785e767d3da1c228
+size 640263