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": 76000.8258379427,
-      "required_pages_per_second_upper_estimate": 0.005257913477680586
+      "required_token_positions_per_second": 75527.92613571833,
+      "required_pages_per_second_upper_estimate": 0.00522519717900222
     },
     {
       "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": 9500.103229742837,
-      "required_pages_per_second_upper_estimate": 0.003977494837578811
+      "required_token_positions_per_second": 9440.990766964791,
+      "required_pages_per_second_upper_estimate": 0.0039527456838222355
     },
     {
       "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": 19000.206459485675,
-      "required_pages_per_second_upper_estimate": 0.003957241936371089
+      "required_token_positions_per_second": 18881.981533929582,
+      "required_pages_per_second_upper_estimate": 0.003932618802178705
     },
     {
       "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": 28500.30968922851,
-      "required_pages_per_second_upper_estimate": 0.001708550048416096
+      "required_token_positions_per_second": 28322.972300894373,
+      "required_pages_per_second_upper_estimate": 0.0016979189427639779
     },
     {
       "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": 28500.30968922851,
-      "required_pages_per_second_upper_estimate": 0.002215773718723606
+      "required_token_positions_per_second": 28322.972300894373,
+      "required_pages_per_second_upper_estimate": 0.002201986516805363
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28500.30968922851,
-      "required_pages_per_second_upper_estimate": 0.3655713714450624
+      "required_token_positions_per_second": 28322.972300894373,
+      "required_pages_per_second_upper_estimate": 0.3632966778375646
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05222524301562714,
-    "producer_p95_seconds": 0.16941408900038368,
-    "gpu_update_seconds": 0.9700947218285619,
-    "producer_batches_per_second": 19.147828564450606,
-    "gpu_batches_per_second": 1.0308271733661931,
-    "producer_headroom": 18.575207424851705,
+    "producer_batch_seconds": 0.3763775794688158,
+    "producer_p95_seconds": 0.17173908699987805,
+    "gpu_update_seconds": 0.9761687335028373,
+    "producer_batches_per_second": 2.6569064007779284,
+    "gpu_batches_per_second": 1.0244130606515616,
+    "producer_headroom": 2.593588956282972,
     "sustainable": true,
-    "buffer_seconds": 60.14587275337084,
+    "buffer_seconds": 60.52246147717591,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 250353.9899390007,
-    "upload_bytes_per_second": 3336063.9758704603,
+    "generation_bytes_per_second": 242621.26608918482,
+    "upload_bytes_per_second": 3845384.767186985,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 53.72013829829708,
+    "estimated_full_shards_per_hour": 52.060875768725,
     "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": [
@@ -220,6 +220,12 @@
       "bytes": 2678432,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.1785862379992977,
+      "bytes": 9722174,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -47,3 +47,4 @@
 {"step": 724700, "loss": 0.5295780301094055, "lr": 5e-05, "grad_norm": 0.21179198168285204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106076008572, "tokens_per_second": 152179.90967603103, "data_wait_s": 1.2849999620812014e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 724800, "loss": 0.5304090976715088, "lr": 5e-05, "grad_norm": 0.21680843165695263, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106090683418, "tokens_per_second": 151308.87995660253, "data_wait_s": 1.2087999493815005e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 724900, "loss": 0.5515720248222351, "lr": 5e-05, "grad_norm": 0.22188084885963635, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106105385064, "tokens_per_second": 145296.97544987826, "data_wait_s": 1.5887999325059354e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 725000, "loss": 0.6052325963973999, "lr": 5e-05, "grad_norm": 0.21800103575122867, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106120080336, "tokens_per_second": 151609.07185543582, "data_wait_s": 2.033100008702604e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ae5cc8af2030d67daf82244aaee23df993e272bde28462923fc229f7aa9a6da8
-size 103947
+oid sha256:532f11033edbb2e54264c6054e69c740a6b4303207e819c41b51fc1c56c1b4d7
+size 106065