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": 76544.67730230112,
-      "required_pages_per_second_upper_estimate": 0.005295538383894157
+      "required_token_positions_per_second": 76574.38069235736,
+      "required_pages_per_second_upper_estimate": 0.005297593333339607
     },
     {
       "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": 9568.08466278764,
-      "required_pages_per_second_upper_estimate": 0.004005957243980918
+      "required_token_positions_per_second": 9571.79758654467,
+      "required_pages_per_second_upper_estimate": 0.004007511767623322
     },
     {
       "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": 19136.16932557528,
-      "required_pages_per_second_upper_estimate": 0.003985559415795655
+      "required_token_positions_per_second": 19143.59517308934,
+      "required_pages_per_second_upper_estimate": 0.003987106024000066
     },
     {
       "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": 28704.25398836292,
-      "required_pages_per_second_upper_estimate": 0.0017207761977442902
+      "required_token_positions_per_second": 28715.39275963401,
+      "required_pages_per_second_upper_estimate": 0.001721443950074073
     },
     {
       "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": 28704.25398836292,
-      "required_pages_per_second_upper_estimate": 0.0022316294909251384
+      "required_token_positions_per_second": 28715.39275963401,
+      "required_pages_per_second_upper_estimate": 0.0022324954813971887
     },
     {
       "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": 28704.25398836292,
-      "required_pages_per_second_upper_estimate": 0.36818734993603114
+      "required_token_positions_per_second": 28715.39275963401,
+      "required_pages_per_second_upper_estimate": 0.3683302261340159
     }
   ],
   "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": 0.41025574878126747,
-    "producer_p95_seconds": 0.21754416700059664,
-    "gpu_update_seconds": 0.9632021794124613,
-    "producer_batches_per_second": 2.437503930098884,
-    "gpu_batches_per_second": 1.0382036309448395,
-    "producer_headroom": 2.3478090977976853,
+    "producer_batch_seconds": 0.05513762575780845,
+    "producer_p95_seconds": 0.1936037369996484,
+    "gpu_update_seconds": 0.9628285509249773,
+    "producer_batches_per_second": 18.136435623697174,
+    "gpu_batches_per_second": 1.0386065089566698,
+    "producer_headroom": 17.462278030508486,
     "sustainable": true,
-    "buffer_seconds": 60.68173730298506,
+    "buffer_seconds": 60.658198708273574,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 293772.40932628955,
+    "generation_bytes_per_second": 293759.6016211004,
     "upload_bytes_per_second": 5601461.319406804,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.03672036973491,
+    "estimated_full_shards_per_hour": 63.033972134349426,
     "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

@@ -102,3 +102,4 @@
 {"step": 614600, "loss": 0.5378708839416504, "lr": 0.0001, "grad_norm": 0.1514035896493702, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89959443471, "tokens_per_second": 138751.9963209251, "data_wait_s": 1.4707000445923768e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 614700, "loss": 0.4195510745048523, "lr": 0.0001, "grad_norm": 0.20096422566688188, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89974139719, "tokens_per_second": 153044.26359626112, "data_wait_s": 1.243699989572633e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 614800, "loss": 0.6015347242355347, "lr": 0.0001, "grad_norm": 0.16205372317747085, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89988811621, "tokens_per_second": 140708.89651951796, "data_wait_s": 1.0165998901356943e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 614900, "loss": 0.5741544961929321, "lr": 0.0001, "grad_norm": 0.15793910587169008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90003501051, "tokens_per_second": 152796.11322089576, "data_wait_s": 1.4789000488235615e-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 sha256:5fb655679b4f72db49342d60f8d8ac7d1235c23451e7da2f1c96783f8f91062f
-size 222003
+oid sha256:2167569ab68d423c04db4123890539324de8849df80066f27789c736b556d9a3
+size 224121