VhddlZddlmZddlmcmZddlmZddgZGddejZGddejZ y)N)Tensor Embedding EmbeddingBagcveZdZdZej Z d dfd ZdefdZ e d dZ dZ xZ S) ra An embedding bag module attached with FakeQuantize modules for weight, used for quantization aware training. We adopt the same interface as `torch.nn.Embedding`, please see https://pytorch.org/docs/stable/generated/torch.nn.Embedding.html#torch.nn.Embedding for documentation. Similar to `torch.nn.Embedding`, with FakeQuantize modules initialized to default. Attributes: weight: fake quant module for weight returnc 8| | d} t |||||||||fi| | sJd| jjtj k(s+Jdt | jjz| |_| j| |_y)Ndevicedtype'qconfig must be provided for QAT moduleSEmbedding weights requires a qscheme of torch.per_channel_affine_float_qparams Got factory_kwargs super__init__weightqschemetorch per_channel_affine_float_qparamsstrqconfigweight_fake_quant)selfnum_embeddings embedding_dim padding_idxmax_norm norm_typescale_grad_by_freqsparse_weightr r rr __class__s U/home/dcms/DCMS/lib/python3.12/site-packages/torch/ao/nn/qat/modules/embedding_ops.pyrzEmbedding.__init__s%+U;           AAAw~~''5+Q+QQ a'.."**+ , Q !(~!Nc tj||j|j|j|j |j |j|jSN) F embeddingrrrrrr r!)rinputs r$forwardzEmbedding.forward>sO{{   " "4;; /    MM NN  # # KK  r%c Lt||jk(s.Jd|jzdz|jjzt|dsJd|jsJd|jj j }|tjk(sJdt|z|j}||j|j|j|j|j|j|j |j | }|S)Create a qat module from a float module Args: `mod` a float module, either produced by torch.ao.quantization utilities or directly from user  qat..from_float only works for r,Input float module must have qconfig defined,Input float module must have a valid qconfigr r)type _FLOAT_MODULE__name__hasattrrrrrrrrrrrrr r!clsmoduse_precomputed_fake_quantweight_qschemerqat_embedding_bags r$ from_floatzEmbedding.from_floatIsCyC--- ll + ,(( ) - sI&V(VV&{{JJJ{++-55!G!GG a.! " G ++       OO LL MM  " " JJ JJ  ! r%c tjj|j|j|j |j |j|j|jd}tjj|jj|_ |j|j|Sr')rnnrrrrrrr r! Parameterrdetachtraintrainingr embedding_bags r$to_floatzEmbedding.to_floatms**          MM NN  # # KK    %xx11$++2D2D2FG DMM*r%) NN@FFNNNNrNF)r5 __module__ __qualname____doc__r?rr4rrr+ classmethodr=rF __classcell__r#s@r$rr sh LLM   O  OD    !!!!F r%c|eZdZdZej Z d dfd Zd defdZ e d dZ dZ xZ S) ra An embedding bag module attached with FakeQuantize modules for weight, used for quantization aware training. We adopt the same interface as `torch.nn.EmbeddingBag`, please see https://pytorch.org/docs/stable/generated/torch.nn.EmbeddingBag.html#torch.nn.EmbeddingBag for documentation. Similar to `torch.nn.EmbeddingBag`, with FakeQuantize modules initialized to default. Attributes: weight: fake quant module for weight rc <| | d}t|||||||||| | f i|| sJd| jjtj k(s+Jdt | jjz| |_| j||_y)Nr r WEmbedding Bag weights requires a qscheme of torch.per_channel_affine_float_qparams Got rr)rrrrrr moder!r"include_last_offsetrrr r rr#s r$rzEmbeddingBag.__init__s %+U;             AAAw~~''5+Q+QQ e'.."**+ , Q !(~!Nr%c tj||j|j||j|j |j |j|j||j|j Sr') r(rErrrrr rSr!rTr)rr*offsetsper_sample_weightss r$r+zEmbeddingBag.forwardse   " "4;; /  MM NN  # # II KK   $ $     r%c xt||jk(s.Jd|jzdz|jjzt|dsJd|jsJd|jj j }|tjk(sJdt|z|j}||j|j|j|j|j|j|j |j |j"|j$| }|S)r-r.r/rr0r1rRr2)r3r4r5r6rrrrrrrrrrr rSr!rTrr7s r$r=zEmbeddingBag.from_floats,CyC--- ll + ,(( ) - sI&V(VV&{{JJJ{++-55!G!GG e.! " G ++       LL MM  " " HH JJ JJ  # # OO  ! r%c tjj|j|j|j |j |j|j|jd|j|j }tjj|jj|_ |j|j |Sr')rr?rrrrrr rSr!rTrr@rrArBrCrDs r$rFzEmbeddingBag.to_floats--       MM NN  # # II KK   $ $      %xx11$++2D2D2FG DMM*r%) NrGFmeanFNFNNNNrH)NNrI)r5rJrKrLr?rr4rrr+rMr=rFrNrOs@r$rr}sn OOM  !$O $OL  v  #!#!Jr%) rtorch.nnr?torch.nn.functional functionalr(r__all__rrr%r$r`sE   'o od{2??{r%