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": 76257.6863536839,
-      "required_pages_per_second_upper_estimate": 0.005275683684158071
+      "required_token_positions_per_second": 77408.60728510693,
+      "required_pages_per_second_upper_estimate": 0.005355307064698402
     },
     {
       "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": 9532.210794210487,
-      "required_pages_per_second_upper_estimate": 0.003990937604339227
+      "required_token_positions_per_second": 9676.075910638367,
+      "required_pages_per_second_upper_estimate": 0.004051170924342325
     },
     {
       "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": 19064.421588420973,
-      "required_pages_per_second_upper_estimate": 0.003970616254261481
+      "required_token_positions_per_second": 19352.151821276733,
+      "required_pages_per_second_upper_estimate": 0.004030542873808827
     },
     {
       "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": 28596.632382631462,
-      "required_pages_per_second_upper_estimate": 0.0017143244468093661
+      "required_token_positions_per_second": 29028.2277319151,
+      "required_pages_per_second_upper_estimate": 0.0017401979289909796
     },
     {
       "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": 28596.632382631462,
-      "required_pages_per_second_upper_estimate": 0.002223262384456933
+      "required_token_positions_per_second": 29028.2277319151,
+      "required_pages_per_second_upper_estimate": 0.002256817024476422
     },
     {
       "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": 28596.632382631462,
-      "required_pages_per_second_upper_estimate": 0.3668068955327851
+      "required_token_positions_per_second": 29028.2277319151,
+      "required_pages_per_second_upper_estimate": 0.3723429372624145
     }
   ],
   "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": 0.13474854691435212,
-    "producer_p95_seconds": 0.32211934899532935,
-    "gpu_update_seconds": 0.9668271295046746,
-    "producer_batches_per_second": 7.421230305627063,
-    "gpu_batches_per_second": 1.034311067080131,
-    "producer_headroom": 7.175046793782513,
+    "producer_batch_seconds": 0.08794624694564845,
+    "producer_p95_seconds": 0.31643289999919944,
+    "gpu_update_seconds": 0.952452221862735,
+    "producer_batches_per_second": 11.37058185800707,
+    "gpu_batches_per_second": 1.0499214312758645,
+    "producer_headroom": 10.82993595453094,
     "sustainable": true,
-    "buffer_seconds": 60.910109158794505,
+    "buffer_seconds": 60.00448997735231,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 523436.2727863979,
-    "upload_bytes_per_second": 4842591.834085885,
+    "generation_bytes_per_second": 523395.2556916616,
+    "upload_bytes_per_second": 4752477.323635754,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 112.31723916715576,
+    "estimated_full_shards_per_hour": 112.30843785345446,
     "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": 1.4460338999997475,
-      "bytes": 11954296,
-      "pages": 39,
-      "shards": 4
-    },
     {
       "seconds": 1.3092962549999356,
       "bytes": 7432740,
@@ -562,6 +556,12 @@
       "bytes": 2714870,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.8995327459997497,
+      "bytes": 6052441,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -339,3 +339,4 @@
 {"step": 958900, "loss": 0.502213716506958, "lr": 3e-05, "grad_norm": 0.3423945327166306, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140356824028, "tokens_per_second": 151072.07346903824, "data_wait_s": 1.2782998965121806e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 959000, "loss": 0.497909814119339, "lr": 3e-05, "grad_norm": 0.3519578454561504, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140371309444, "tokens_per_second": 150723.64260279687, "data_wait_s": 9.571995178703219e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 959100, "loss": 0.5026146173477173, "lr": 3e-05, "grad_norm": 0.3385202518930032, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140386033764, "tokens_per_second": 153723.02122442852, "data_wait_s": 1.0247997124679387e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 959200, "loss": 0.5705493092536926, "lr": 3e-05, "grad_norm": 0.3536448166679467, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140400686195, "tokens_per_second": 152233.72911656197, "data_wait_s": 1.6534002497792244e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c7b90dec2ed8ba7e9352d520dc82b79d2e4f288ce361f790008c969b7c1de579
-size 727101
+oid sha256:9e5c44f484f61da4514b2999af9eb6c6237f866b765a1d2624b53c780512bc4b
+size 729219