tj$ddlmZddlmZmZddlZddlZddlm Z ddlm Z m Z d'd Z d(d Z d'd Zd)dZed*dZed+dZd,dZed-dZed.dZed/dZd0dZd1d"Zd2d%Zd3d&ZdS)4) annotations)AnyoverloadN) coo_matrix)Tensordevicealist | np.ndarray | Tensorreturnrcht|trLtd|Drtjd|DStj|}n)t|t stj|}|jr |tj S|S)a Converts the input `a` to a PyTorch tensor if it is not already a tensor. Handles lists of sparse tensors by stacking them. Args: a (Union[list, np.ndarray, Tensor]): The input array or tensor. Returns: Tensor: The converted tensor. c3NK|] }t|to|jV!dSN) isinstancer is_sparse.0xs _/var/www/html/CA-Chatbot/venv/lib/python3.11/site-packages/sentence_transformers/util/tensor.py z%_convert_to_tensor..s3@@z!V$$4@@@@@@crg|]4}|tj5S)dtype)coalescetotorchfloat32rs r z&_convert_to_tensor..s0PPP em D DPPPrr) rlistallrstacktensorrrrrr s r_convert_to_tensorr$ s!T @@a@@@ @ @ ;PPaPPPQQ Q QAA 6 " " LOO{)tt%-t((( Hrc`|dkr|d}|S)z If the tensor `a` is 1-dimensional, it is unsqueezed to add a batch dimension. Args: a (Tensor): The input tensor. Returns: Tensor: The tensor with a batch dimension. r)dim unsqueezer#s r_convert_to_batchr)$s) uuww!|| KKNN Hrc~t|}|dkr|d}|S)a Converts the input data to a tensor with a batch dimension. Handles lists of sparse tensors by stacking them. Args: a (Union[list, np.ndarray, Tensor]): The input data to be converted. Returns: Tensor: The converted tensor with a batch dimension. r&r)r$r'r(r#s r_convert_to_batch_tensorr+3s6 1Auuww!|| KKNN Hr embeddingsc~|js'tjj|ddS|}||}}tj| d|j }| d|d|dztj | d|d}|dk}|}||xx||zcc<tj||| S)z Normalizes the embeddings matrix, so that each sentence embedding has unit length. Args: embeddings (Tensor): The input embeddings matrix. Returns: Tensor: The normalized embeddings matrix. r&)pr'r)r)rrnn functional normalizerindicesvalueszerossizer index_add_sqrt index_selectclonesparse_coo_tensor)r,r3r4 row_normsmasknormalized_valuess rnormalize_embeddingsr?Ds#  Ex",,Z1!,DDD$$&&J ((**J,=,=,?,?VG JOOA..z7HIIII GAJ 222 9%%221gajAAI q=D dy.  "7,=z?P?P Q QQr np.ndarray truncate_dim int | NonecdSrr,rAs rtruncate_embeddingsrFaY\Y\r torch.TensorcdSrrDrEs rrFrFe]`]`rnp.ndarray | torch.Tensorc|dd|fS)a Truncates the embeddings matrix. Args: embeddings (Union[np.ndarray, torch.Tensor]): Embeddings to truncate. truncate_dim (Optional[int]): The dimension to truncate sentence embeddings to. `None` does no truncation. Example: >>> from sentence_transformers import SentenceTransformer >>> from sentence_transformers.util import truncate_embeddings >>> model = SentenceTransformer("tomaarsen/mpnet-base-nli-matryoshka") >>> embeddings = model.encode(["It's so nice outside!", "Today is a beautiful day.", "He drove to work earlier"]) >>> embeddings.shape (3, 768) >>> model.similarity(embeddings, embeddings) tensor([[1.0000, 0.8100, 0.1426], [0.8100, 1.0000, 0.2121], [0.1426, 0.2121, 1.0000]]) >>> truncated_embeddings = truncate_embeddings(embeddings, 128) >>> truncated_embeddings.shape >>> model.similarity(truncated_embeddings, truncated_embeddings) tensor([[1.0000, 0.8092, 0.1987], [0.8092, 1.0000, 0.2716], [0.1987, 0.2716, 1.0000]]) Returns: Union[np.ndarray, torch.Tensor]: Truncated embeddings. .NrDrEs rrFrFis: c=L=( ))rmax_active_dimsintcdSrrDr,rMs rselect_max_active_dimsrQsililrNonecdSrrDrPs rrQrQrGrcdSrrDrPs rrQrQrJrc||S|dkrtd|dt|tjrt j|}|jd}t jt j|t||d\}}t j |}| d|| d||S)a Returns a new tensor with only the top-k values (in absolute terms) of each embedding, all others set to zero. The input embeddings are never modified in place. Args: embeddings (Union[np.ndarray, torch.Tensor]): Embeddings to sparsify by keeping only the largest values. max_active_dims (Optional[int]): Number of values to keep as non-zeros per embedding. `None` keeps all values, returning the embeddings as-is. Raises: ValueError: If `max_active_dims` is 0 or negative. Returns: Union[np.ndarray, torch.Tensor]: A new dense tensor of the same shape, with all but the top-k values per embedding set to zero. If `max_active_dims` is `None`, the embeddings are returned unchanged. Nrz0max_active_dims must be a positive integer, got .)kr') ValueErrorrnpndarrayrr"shapetopkabsmin zeros_likescatter_gather)r,rM embedding_dim_ top_indicesselecteds rrQrQs(!^O^^^___*bj)).\*-- $R(MZ * 5 5_m9\9\bdeeeNA{ ++H b+z'8'8['I'IJJJ Orbatchdict[str, Any] target_devicerc|D];}t||tr|||||<<|S)au Send a PyTorch batch (i.e., a dictionary of string keys to Tensors) to a device (e.g. "cpu", "cuda", "mps"). Args: batch (Dict[str, Tensor]): The batch to send to the device. target_device (torch.device): The target device (e.g. "cpu", "cuda", "mps"). Returns: Dict[str, Tensor]: The batch with tensors sent to the target device. )rrr)rgrikeys rbatch_to_devicerlsH66 eCj& ) ) 6s}55E#J LrrrcV|}|}|}t ||d|dff|jS)ac Converts a sparse PyTorch tensor to a SciPy COO sparse matrix. Args: x (torch.Tensor): A 2-dimensional sparse PyTorch tensor in COO layout. Example: >>> import torch >>> from sentence_transformers.util import to_scipy_coo >>> x = torch.sparse_coo_tensor([[0, 1], [1, 0]], [1.0, 2.0], (3, 3)) >>> to_scipy_coo(x).toarray() array([[0., 1., 0.], [2., 0., 0.], [0., 0., 0.]], dtype=float32) Returns: scipy.sparse.coo_matrix: A SciPy COO matrix with the same shape, indices and values as the input tensor. rr&)r\)rr3cpunumpyr4rr\)rr3r4s r to_scipy_coorps~& Aiikkoo%%''G XXZZ^^   # # % %F v GAJ78 H H HHrc|js|}|}tj|d|jtj}|dkr+d|| <|S|dkrv| dkrJ| }tj |dd\}}|||<|Std|d ) a Compute count vector from sparse embeddings indicating how many samples have non-zero values in each dimension. Args: embeddings: Sparse tensor of shape (batch_size, vocab_size) or (vocab_size,) Returns: Count vector of shape (vocab_size,) rW)rrr&r.rT) return_countszExpected 1D or 2D tensor, got D)r to_sparserrr5r6rint32r'r3squeezer4numeluniquerNrY)r, count_vectorr3 unique_dimscountss rcompute_count_vectorr|s;  ,))++ $$&&J;zr22:;LTYT_```L~~178 Z''))11334   Q        $ $ & & * * ((**G"',wqz"N"N"N K(. L %M*..:J:JMMMNNNr)r r r r)r rr r)r,rr r)r,r@rArBr r@)r,rHrArBr rH)r,rKrArBr rK)r,rKrMrNr rH)r,r@rMrRr r@)r,rHrMrRr rH)r,rKrMrBr rK)rgrhrirr rh)rrr r)r,rHr rH) __future__rtypingrrrorZr scipy.sparserrrr$r)r+r?rFrQrlrpr|rDrrrs""""""  ######     2         "RRRR: \\\ \ ``` `****@ lll l \\\ \ ``` `%%%%P"IIII2OOOOOOr