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": 77080.12649840051,
-      "required_pages_per_second_upper_estimate": 0.005332582001693618
+      "required_token_positions_per_second": 76703.48929297118,
+      "required_pages_per_second_upper_estimate": 0.005306525365903298
     },
     {
       "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": 9635.015812300064,
-      "required_pages_per_second_upper_estimate": 0.004033979918600433
+      "required_token_positions_per_second": 9587.936161621397,
+      "required_pages_per_second_upper_estimate": 0.004014268651996177
     },
     {
       "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": 19270.031624600128,
-      "required_pages_per_second_upper_estimate": 0.004013439402496311
+      "required_token_positions_per_second": 19175.872323242795,
+      "required_pages_per_second_upper_estimate": 0.00399382850317134
     },
     {
       "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": 28905.04743690019,
-      "required_pages_per_second_upper_estimate": 0.0017328134583902558
+      "required_token_positions_per_second": 28763.808484864192,
+      "required_pages_per_second_upper_estimate": 0.0017243463988750904
     },
     {
       "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": 28905.04743690019,
-      "required_pages_per_second_upper_estimate": 0.0022472402983518703
+      "required_token_positions_per_second": 28763.808484864192,
+      "required_pages_per_second_upper_estimate": 0.002236259591075566
     },
     {
       "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": 28905.04743690019,
-      "required_pages_per_second_upper_estimate": 0.37076291269865946
+      "required_token_positions_per_second": 28763.808484864192,
+      "required_pages_per_second_upper_estimate": 0.3689512510725131
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.31896324820317545,
-    "producer_p95_seconds": 0.3049363850004738,
-    "gpu_update_seconds": 0.9565111443029344,
-    "producer_batches_per_second": 3.1351574378343834,
-    "gpu_batches_per_second": 1.0454661254665867,
-    "producer_headroom": 2.9988130284328216,
+    "producer_batch_seconds": 0.4792551796250848,
+    "producer_p95_seconds": 0.3039746210015437,
+    "gpu_update_seconds": 0.9612079017473871,
+    "producer_batches_per_second": 2.086571084703325,
+    "gpu_batches_per_second": 1.040357656425933,
+    "producer_headroom": 2.0056286141744524,
     "sustainable": true,
-    "buffer_seconds": 60.260202091084864,
+    "buffer_seconds": 60.55609781008539,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 342767.96064615325,
-    "upload_bytes_per_second": 5355644.6682092,
+    "generation_bytes_per_second": 309682.09046949976,
+    "upload_bytes_per_second": 5342353.476098344,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.55002512491653,
+    "estimated_full_shards_per_hour": 66.4505675846457,
     "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": 1.198957466000138,
-      "bytes": 7513050,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 3.0025762599980226,
       "bytes": 963365,
@@ -562,6 +556,12 @@
       "bytes": 11997417,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.3358469599988894,
+      "bytes": 7186422,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -312,3 +312,4 @@
 {"step": 635600, "loss": 0.538130521774292, "lr": 0.0001, "grad_norm": 0.16519310237239834, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93034955828, "tokens_per_second": 152800.79926206343, "data_wait_s": 1.4249002560973167e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 635700, "loss": 0.578747570514679, "lr": 0.0001, "grad_norm": 0.17352032128901876, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93049641333, "tokens_per_second": 152073.30066789777, "data_wait_s": 7.675000233575702e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 635800, "loss": 0.6068922877311707, "lr": 0.0001, "grad_norm": 0.21987812565563367, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93064271505, "tokens_per_second": 152040.99320863085, "data_wait_s": 1.4780998753849417e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 635900, "loss": 0.5734906196594238, "lr": 0.0001, "grad_norm": 0.1751389047712576, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93079051902, "tokens_per_second": 131361.38319660482, "data_wait_s": 1.3438999303616583e-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 sha256:8245a3749d4bb2547254fb53f86c5b313471b68772c9cbef31e3a33ac5c0b51a
-size 670698
+oid sha256:2b264a040b3112f2dabc7e54c71d3f67b0a3a7c31571e1632aead7a4e992b7ef
+size 672816