| 1 | # SPDX-License-Identifier: Apache-2.0 |
| 2 | # SPDX-FileCopyrightText: Copyright contributors to the vLLM project |
| 3 | |
| 4 | import pytest |
| 5 | |
| 6 | from tests.models.language.pooling.embed_utils import correctness_test_embed_models |
| 7 | from tests.models.utils import EmbedModelInfo |
| 8 | |
| 9 | from .mteb_embed_utils import mteb_test_embed_models |
| 10 | |
| 11 | MODELS = [ |
| 12 | EmbedModelInfo( |
| 13 | "nomic-ai/nomic-embed-text-v1", |
| 14 | architecture="NomicBertModel", |
| 15 | # Fixme: |
| 16 | # Update nomic-embed code to support the latest |
| 17 | # HF version and remove revision set. |
| 18 | revision="720244025c1a7e15661a174c63cce63c8218e52b", |
| 19 | mteb_score=0.737568559, |
| 20 | mteb_tol=2e-3, |
| 21 | enable_test=True, |
| 22 | seq_pooling_type="MEAN", |
| 23 | attn_type="encoder_only", |
| 24 | is_prefix_caching_supported=False, |
| 25 | is_chunked_prefill_supported=False, |
| 26 | ), |
| 27 | EmbedModelInfo( |
| 28 | "nomic-ai/nomic-embed-text-v1.5", |
| 29 | architecture="NomicBertModel", |
| 30 | enable_test=False, |
| 31 | ), |
| 32 | EmbedModelInfo( |
| 33 | "nomic-ai/CodeRankEmbed", architecture="NomicBertModel", enable_test=False |
| 34 | ), |
| 35 | EmbedModelInfo( |
| 36 | "nomic-ai/nomic-embed-text-v2-moe", |
| 37 | architecture="NomicBertModel", |
| 38 | mteb_score=0.715488912, |
| 39 | mteb_tol=2e-3, |
| 40 | enable_test=True, |
| 41 | seq_pooling_type="MEAN", |
| 42 | attn_type="encoder_only", |
| 43 | is_prefix_caching_supported=False, |
| 44 | is_chunked_prefill_supported=False, |
| 45 | ), |
| 46 | ] |
| 47 | |
| 48 | |
| 49 | @pytest.mark.flaky(reruns=2) |
| 50 | @pytest.mark.parametrize("model_info", MODELS) |
| 51 | def test_embed_models_mteb(hf_runner, vllm_runner, model_info: EmbedModelInfo) -> None: |
| 52 | mteb_test_embed_models(hf_runner, vllm_runner, model_info) |
| 53 | |
| 54 | |
| 55 | @pytest.mark.parametrize("model_info", MODELS) |
| 56 | def test_embed_models_correctness( |
| 57 | hf_runner, vllm_runner, model_info: EmbedModelInfo, example_prompts |
| 58 | ) -> None: |
| 59 | correctness_test_embed_models(hf_runner, vllm_runner, model_info, example_prompts) |