diff --git a/extra/datasets/sops.gz b/extra/datasets/sops.gz
index 1362cbe16d..9b61b87ad2 100644
Binary files a/extra/datasets/sops.gz and b/extra/datasets/sops.gz differ
diff --git a/extra/optimization/generate_dataset.sh b/extra/optimization/generate_dataset.sh
index e4a7fe03d1..5413d56ab1 100755
--- a/extra/optimization/generate_dataset.sh
+++ b/extra/optimization/generate_dataset.sh
@@ -6,24 +6,12 @@ export CAPTURE_PROCESS_REPLAY=1
 rm $LOGOPS
 test/external/process_replay/reset.py
 
-python3 -m pytest -n=auto test/ --ignore=test/unit --durations=20
-STEPS=3 python3 examples/hlb_cifar10.py
-WINO=1 STEPS=3 python3 examples/hlb_cifar10.py
-python3 examples/stable_diffusion.py --noshow
-python3 examples/llama.py --prompt "hello" --count 5
-python3 examples/gpt2.py --count 5
-HALF=1 python3 examples/gpt2.py --count 5
-python3 examples/beautiful_mnist.py
-python3 examples/beautiful_cartpole.py
-python3 examples/mlperf/model_spec.py
-python3 examples/yolov8.py ./test/models/efficientnet/Chicken.jpg
-examples/openpilot/go.sh
-JIT=2 BIG=1 MPS=1 pytest -n=auto test/ --ignore=test/test_fusion_op.py --ignore=test/test_gc.py --ignore=test/test_speed_v_torch.py --ignore=test/test_jit.py
-JIT=2 BIG=1 MPS=1 python -m pytest test/test_gc.py
-JIT=2 BIG=1 MPS=1 python -m pytest test/test_jit.py
-JIT=2 BIG=1 MPS=1 python -m pytest test/test_speed_v_torch.py
+CI=1 python3 -m pytest -n=auto test/test_ops.py test/test_nn.py test/test_winograd.py test/models/test_real_world.py --durations=20
+GPU=1 python3 -m pytest test/test_tiny.py
 
 # extract, sort and uniq
 extra/optimization/extract_dataset.py
 sort -u /tmp/ops > /tmp/sops
 ls -lh /tmp/ops /tmp/sops
+# gzip -k /tmp/sops
+# mv /tmp/sops.gz extra/datasets/
\ No newline at end of file
diff --git a/extra/optimization/helpers.py b/extra/optimization/helpers.py
index 4d2549dee6..89ae8eccdd 100644
--- a/extra/optimization/helpers.py
+++ b/extra/optimization/helpers.py
@@ -27,7 +27,7 @@ from tinygrad.helpers import dedup, DEBUG
 def load_worlds(filter_reduce=True, filter_noimage=True, filter_novariable=True):
   fn = Path(__file__).parent.parent / "datasets/sops.gz"
   ast_strs = dedup(gzip.open(fn).read().decode('utf-8').strip().split("\n"))
-  assert len(ast_strs) >= getenv("MIN_ASTS", 5000), f"dataset size = {len(ast_strs)} is too small"
+  assert len(ast_strs) >= getenv("MIN_ASTS", 1000), f"dataset size = {len(ast_strs)} is too small"
   if DEBUG >= 1: print(f"loaded {len(ast_strs)=} before filters")
   if filter_reduce: ast_strs = [x for x in ast_strs if "REDUCE_AXIS" in x]
   if filter_noimage: ast_strs = [x for x in ast_strs if "dtypes.image" not in x]