a h@sddlmZddlZddlZddlmZddlmZddlm Z ddl m Z m Z m Z mZe eZdddd d d Zd d dd d dZdddd ddZd d dd ddZdddd ddZd d dd ddZd d dddZd d dd ddZd d dddZdddddd ZdS)!) annotationsN)pairwise_distances)Tensor)logging)_convert_to_batch_tensor_convert_to_tensornormalize_embeddings to_scipy_coor)abreturncCs t||S) Computes the cosine similarity between two tensors. Args: a (Union[list, np.ndarray, Tensor]): The first tensor. b (Union[list, np.ndarray, Tensor]): The second tensor. Returns: Tensor: Matrix with res[i][j] = cos_sim(a[i], b[j]) )cos_simr r ra/var/www/html/assistant/venv/lib/python3.9/site-packages/sentence_transformers/util/similarity.pypytorch_cos_sims rzlist | np.ndarray | TensorcCs8t|}t|}t|}t|}t||ddS)rrr)rr torchmm transposeto_denser r Za_normZb_normrrrrs rcCsZt|}t|}|js|jr@t|}t|}||jddStt|t|SdS)a Computes the pairwise cosine similarity cos_sim(a[i], b[i]). Args: a (Union[list, np.ndarray, Tensor]): The first tensor. b (Union[list, np.ndarray, Tensor]): The second tensor. Returns: Tensor: Vector with res[i] = cos_sim(a[i], b[i]) dimN)r is_sparser sumrpairwise_dot_scorerrrrpairwise_cos_sim0s  rcCs(t|}t|}t||ddS)a Computes the dot-product dot_prod(a[i], b[j]) for all i and j. Args: a (Union[list, np.ndarray, Tensor]): The first tensor. b (Union[list, np.ndarray, Tensor]): The second tensor. Returns: Tensor: Matrix with res[i][j] = dot_prod(a[i], b[j]) rr)rrrrrrrrr dot_scoreGs r cCs$t|}t|}||jddS)a Computes the pairwise dot-product dot_prod(a[i], b[i]). Args: a (Union[list, np.ndarray, Tensor]): The first tensor. b (Union[list, np.ndarray, Tensor]): The second tensor. Returns: Tensor: Vector with res[i] = dot_prod(a[i], b[i]) rr)rrrrrrrrXs rcCszt|}t|}|js|jr`tdt|}t|}t||dd}t|  |j  Stj ||dd SdS)a Computes the manhattan similarity (i.e., negative distance) between two tensors. Handles sparse tensors without converting to dense when possible. Args: a (Union[list, np.ndarray, Tensor]): The first tensor. b (Union[list, np.ndarray, Tensor]): The second tensor. Returns: Tensor: Matrix with res[i][j] = -manhattan_distance(a[i], b[j]) z8Using scipy for sparse Manhattan similarity computation.Z manhattan)Zmetricg?pN) rrlogger warning_oncer rrZ from_numpyfloattoZdevicercdist)r r Za_cooZb_coodistrrr manhattan_simis   r)rcCs.t|}t|}tjt||dd S)a< Computes the manhattan similarity (i.e., negative distance) between pairs of tensors. Args: a (Union[list, np.ndarray, Tensor]): The first tensor. b (Union[list, np.ndarray, Tensor]): The second tensor. Returns: Tensor: Vector with res[i] = -manhattan_distance(a[i], b[i]) rr)rrrabsrrrrrpairwise_manhattan_sims r+cCst|}t|}|jrtjj||ddd}tjj||ddd}t||}|d||}tj |dd}t | Stj ||dd Sd S) a Computes the euclidean similarity (i.e., negative distance) between two tensors. Handles sparse tensors without converting to dense when possible. Args: a (Union[list, np.ndarray, Tensor]): The first tensor. b (Union[list, np.ndarray, Tensor]): The second tensor. Returns: Tensor: Matrix with res[i][j] = -euclidean_distance(a[i], b[j]) rrrg)ming@r!N) rrrsparserrZ unsqueezematmultclampsqrtr')r r Z a_norm_sqZ b_norm_sqZ dot_productZ squared_distrrr euclidean_sims r3cCs2t|}t|}ttj||ddd S)a: Computes the euclidean distance (i.e., negative distance) between pairs of tensors. Args: a (Union[list, np.ndarray, Tensor]): The first tensor. b (Union[list, np.ndarray, Tensor]): The second tensor. Returns: Tensor: Vector with res[i] = -euclidean_distance(a[i], b[i]) r,rr)rrr2rrrrrrpairwise_euclidean_sims r4)xyr c Cs"|jr(td|}|}t|}t|}tj|ddd\}}tj|ddd\}}tj|d|dddd}|||||}|||||}tj|d|ddddd} tj|d|ddddd} || | }|| | }tjtj ||fdddd} t | S)aP Computes the absolute normalized angle distance. See :class:`~sentence_transformers.losses.AnglELoss` or https://arxiv.org/abs/2309.12871v1 for more information. Args: x (Tensor): The first tensor. y (Tensor): The second tensor. Returns: Tensor: Vector with res[i] = angle_sim(a[i], b[i]) zOPairwise angle similarity does not support sparse tensors. Converting to dense.r,rrT)rZkeepdimg?) rr#r$Zcoalescerrrchunkrconcatr*) r5r6r r cdzreZimZdzZdwZ norm_anglerrrpairwise_angle_sims"        r=) __future__rnumpynprZsklearn.metricsrrZtransformers.utilsrZtensorrrr r Z get_logger__name__r#rrrr rr)r+r3r4r=rrrrs"