Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.
Python agentaichatgptdeepseekfine-tuninggemmaimage-generationllamallmllmsopenaipythonqwenreinforcement-learningself-hostedstable-diffusiontext-to-speechttsuiunsloth
RAG ANN test: build the recall check index on one thread (#16308)
Daniel Han5cbd693a247a
diff --git a/studio/backend/tests/test_rag_vec_ann.py b/studio/backend/tests/test_rag_vec_ann.pyindex f642e68168b..ce596050882 100644--- a/studio/backend/tests/test_rag_vec_ann.py+++ b/studio/backend/tests/test_rag_vec_ann.py@@ -36,9 +36,13 @@ def _exact(rows, keys, q, k):return [int(keys[i]) for i in order]-def _built(rows, keys):+def _built(+ rows,+ keys,+ threads = 2,+):index = vec_ann.new_index(rows.shape[1])- vec_ann.add(np, index, keys, rows, threads = 2)+ vec_ann.add(np, index, keys, rows, threads = threads)index.expansion_search = vec_ann.EXPANSION_SEARCHempty = np.empty(0, np.int64), np.empty((0, rows.shape[1]), np.float32)return vec_ann.Ann(index, len(index), None, *empty)@@ -47,7 +51,9 @@ def _built(rows, keys):def test_search_recalls_the_exact_neighbours():rows = _clustered(5000, 32)keys = np.arange(5000) * 3 + 7- ann = _built(rows, keys)+ # One thread: a parallel build's graph depends on how the threads interleave, and read 0.983 on a CI+ # runner. The bar is for the index and its search, not for the scheduler.+ ann = _built(rows, keys, threads = 1)rng = np.random.default_rng(1)hits = 0for q in rows[rng.integers(0, 5000, 20)] + 0.01: