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": 76957.01079124212,
-      "required_pages_per_second_upper_estimate": 0.005324064571404603
+      "required_token_positions_per_second": 76501.31462938494,
+      "required_pages_per_second_upper_estimate": 0.005292538453566552
     },
     {
       "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": 9619.626348905265,
-      "required_pages_per_second_upper_estimate": 0.00402753667164557
+      "required_token_positions_per_second": 9562.664328673118,
+      "required_pages_per_second_upper_estimate": 0.004003687863276616
     },
     {
       "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": 19239.25269781053,
-      "required_pages_per_second_upper_estimate": 0.004007028963740921
+      "required_token_positions_per_second": 19125.328657346236,
+      "required_pages_per_second_upper_estimate": 0.003983301590491191
     },
     {
       "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": 28858.879046715796,
-      "required_pages_per_second_upper_estimate": 0.0017300457338937515
+      "required_token_positions_per_second": 28687.992986019355,
+      "required_pages_per_second_upper_estimate": 0.0017198013754829062
     },
     {
       "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": 28858.879046715796,
-      "required_pages_per_second_upper_estimate": 0.0022436509090883205
+      "required_token_positions_per_second": 28687.992986019355,
+      "required_pages_per_second_upper_estimate": 0.0022303652695175166
     },
     {
       "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": 28858.879046715796,
-      "required_pages_per_second_upper_estimate": 0.37017071416112923
+      "required_token_positions_per_second": 28687.992986019355,
+      "required_pages_per_second_upper_estimate": 0.36797877125767187
     }
   ],
   "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": 0.4772530617109112,
-    "producer_p95_seconds": 0.2727155809989199,
-    "gpu_update_seconds": 0.958041369356181,
-    "producer_batches_per_second": 2.095324431057783,
-    "gpu_batches_per_second": 1.0437962618169776,
-    "producer_headroom": 2.007407487176059,
+    "producer_batch_seconds": 0.23744671430458197,
+    "producer_p95_seconds": 0.27662627800236805,
+    "gpu_update_seconds": 0.9637481441616993,
+    "producer_batches_per_second": 4.211471204934265,
+    "gpu_batches_per_second": 1.0376154870522047,
+    "producer_headroom": 4.058797557945834,
     "sustainable": true,
-    "buffer_seconds": 60.3566062694394,
+    "buffer_seconds": 60.716133082187056,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 378183.77904573246,
-    "upload_bytes_per_second": 5004492.974510393,
+    "generation_bytes_per_second": 366159.86989718094,
+    "upload_bytes_per_second": 5062978.411259473,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.14943531540852,
+    "estimated_full_shards_per_hour": 78.56938431440898,
     "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": 1.1097759929980384,
-      "bytes": 7296844,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.8535761809980613,
       "bytes": 4986805,
@@ -562,6 +556,12 @@
       "bytes": 2165474,
       "pages": 22,
       "shards": 2
+    },
+    {
+      "seconds": 1.0913281839966658,
+      "bytes": 12193503,
+      "pages": 46,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -252,3 +252,4 @@
 {"step": 629600, "loss": 0.5247787833213806, "lr": 0.0001, "grad_norm": 0.15267186663726345, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92156825405, "tokens_per_second": 152580.60101235227, "data_wait_s": 8.753002475714311e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 629700, "loss": 0.546468198299408, "lr": 0.0001, "grad_norm": 0.17180763845681174, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92171527225, "tokens_per_second": 152272.4774708481, "data_wait_s": 1.619900285732001e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 629800, "loss": 0.5240869522094727, "lr": 0.0001, "grad_norm": 0.1598519017610291, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92186182462, "tokens_per_second": 130941.71611236915, "data_wait_s": 1.3990997103974223e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 629900, "loss": 0.5332528352737427, "lr": 0.0001, "grad_norm": 0.18037494787358488, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92200992448, "tokens_per_second": 153271.32318847603, "data_wait_s": 1.9612001779023558e-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 sha256:1aa2518f8772d85bc594ceae8cd51f3574b48b6497c3ce1485f2ee142af10d7e
-size 542052
+oid sha256:d2b4375bf1fa77cd42069ac5abc954529160a6d3904d94dd7aff0c4300ad4ebf
+size 544170