fix(python): support deterministic IVF-PQ training

python/python/lancedb/index.py

@@ -769,6 +769,11 @@ class IvfPq:
 
         The default value is 256.
 
+    seed: int, optional
+        Seed used for deterministic sampling and training. Given identical data in
+        the same row order and identical index parameters, the same seed produces
+        the same IVF centroids and PQ codebook. If omitted, training remains random.
+
     target_partition_size: int, default is 8192
 
         The target size of each partition.
@@ -783,6 +788,7 @@ class IvfPq:
     num_bits: int = 8
     max_iterations: int = 50
     sample_rate: int = 256
+    seed: Optional[int] = None
     target_partition_size: Optional[int] = None
     # Name of the accelerator (e.g. "cuda") to use for IVF training. When set,
     # create_index() dispatches to pylance to build the index on the accelerator.

python/python/tests/test_index.py

@@ -375,7 +375,7 @@ async def test_create_vector_index(some_table: AsyncTable):
 @pytest.mark.asyncio
 async def test_create_4bit_ivfpq_index(some_table: AsyncTable):
     # Can create
-    await some_table.create_index("vector", config=IvfPq(num_bits=4))
+    await some_table.create_index("vector", config=IvfPq(num_bits=4, seed=42))
     # Can recreate if replace=True
     await some_table.create_index("vector", config=IvfPq(num_bits=4), replace=True)
     # Can't recreate if replace=False

python/src/index.rs

@@ -90,6 +90,9 @@ pub fn extract_index_params(source: &Option<Bound<'_, PyAny>>) -> PyResult<Lance
                     .max_iterations(params.max_iterations)
                     .sample_rate(params.sample_rate)
                     .num_bits(params.num_bits);
+                if let Some(seed) = params.seed {
+                    ivf_pq_builder = ivf_pq_builder.seed(seed);
+                }
                 if let Some(num_partitions) = params.num_partitions {
                     ivf_pq_builder = ivf_pq_builder.num_partitions(num_partitions);
                 }
@@ -232,6 +235,7 @@ struct IvfPqParams {
     num_bits: u32,
     max_iterations: u32,
     sample_rate: u32,
+    seed: Option<u64>,
     target_partition_size: Option<u32>,
 }