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": 77124.13977300709,
-      "required_pages_per_second_upper_estimate": 0.005335626942155769
+      "required_token_positions_per_second": 77257.12951703987,
+      "required_pages_per_second_upper_estimate": 0.0053448274812266765
     },
     {
       "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": 9640.517471625886,
-      "required_pages_per_second_upper_estimate": 0.0040362833484724435
+      "required_token_positions_per_second": 9657.141189629983,
+      "required_pages_per_second_upper_estimate": 0.004043243352058052
     },
     {
       "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": 19281.034943251772,
-      "required_pages_per_second_upper_estimate": 0.004015731103594422
+      "required_token_positions_per_second": 19314.282379259967,
+      "required_pages_per_second_upper_estimate": 0.00402265566772108
     },
     {
       "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": 28921.552414877657,
-      "required_pages_per_second_upper_estimate": 0.0017338029066183628
+      "required_token_positions_per_second": 28971.42356888995,
+      "required_pages_per_second_upper_estimate": 0.0017367926061525568
     },
     {
       "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": 28921.552414877657,
-      "required_pages_per_second_upper_estimate": 0.0022485234877918925
+      "required_token_positions_per_second": 28971.42356888995,
+      "required_pages_per_second_upper_estimate": 0.0022524007506563243
     },
     {
       "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": 28921.552414877657,
-      "required_pages_per_second_upper_estimate": 0.3709746208344898
+      "required_token_positions_per_second": 28971.42356888995,
+      "required_pages_per_second_upper_estimate": 0.3716143144506862
     }
   ],
   "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": 0.1544521663828391,
-    "producer_p95_seconds": 0.3489458390031359,
-    "gpu_update_seconds": 0.955965281648487,
-    "producer_batches_per_second": 6.474496430961737,
-    "gpu_batches_per_second": 1.046063093709406,
-    "producer_headroom": 6.1893938041564605,
+    "producer_batch_seconds": 0.09514422385950638,
+    "producer_p95_seconds": 0.3557029380026506,
+    "gpu_update_seconds": 0.9543196914109852,
+    "producer_batches_per_second": 10.510359530355078,
+    "gpu_batches_per_second": 1.0478668825553368,
+    "producer_headroom": 10.030243063626965,
     "sustainable": true,
-    "buffer_seconds": 60.22581274385468,
+    "buffer_seconds": 60.12214055889206,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 348821.8071243641,
-    "upload_bytes_per_second": 4750578.655387151,
+    "generation_bytes_per_second": 349178.9782507177,
+    "upload_bytes_per_second": 4702680.715586682,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.84903965280715,
+    "estimated_full_shards_per_hour": 74.92568026200435,
     "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": 1.4301322219980648,
-      "bytes": 6037828,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 1.568642534999526,
       "bytes": 11958930,
@@ -562,6 +556,12 @@
       "bytes": 7690785,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.3696625419979682,
+      "bytes": 1334642,
+      "pages": 36,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -381,3 +381,4 @@
 {"step": 642500, "loss": 0.5789816975593567, "lr": 0.0001, "grad_norm": 0.17352284766246504, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94045257152, "tokens_per_second": 152735.16706581277, "data_wait_s": 1.528499706182629e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 642600, "loss": 0.587306022644043, "lr": 0.0001, "grad_norm": 0.17588659467996431, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94060057780, "tokens_per_second": 153534.6000730757, "data_wait_s": 1.1534997611306608e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 642700, "loss": 0.506992518901825, "lr": 0.0001, "grad_norm": 0.17536772722896798, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94074894927, "tokens_per_second": 154416.61329765926, "data_wait_s": 1.3270000636111945e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 642800, "loss": 0.5037686824798584, "lr": 0.0001, "grad_norm": 0.16527506314839135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94089365992, "tokens_per_second": 143680.36548489326, "data_wait_s": 1.077799970516935e-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 sha256:1066c4b02af4cc5f9a53f2819a85bdb457d0f65858814b0f9cdaa0d1088749f1
-size 817623
+oid sha256:21f14b0ca319849b03d3766e7c3c2d5b0ae55c5ba37a4264e5926a1da358354b
+size 819741