tjddlZddlmZejeZ ddejdejdejd ejd ejd ed eed e de dejfdZ ddZ dS)N)loggingmean token_logitsduration_logitstargets logit_lengthstarget_lengthsblank_token_id durationssigma reductionreturnc Z d} || vr3td|ddd| Dd|j} |j\} } } }|}|}t j|d|z }t j|d}t j| | | ft d | }d |d d d d f<|d d d d d d |f}| dkr||d d| d}t j |d d d d d | dz d d fd|d d}t j t d | }td| | zdz D]}td || z dz}t|dz| }t j||| }||z }g}t#|D]k\}}||z }|d k}|s&|d } |d krh|d d | |f|d d | |fz|d d | ||fz}!t j|d |!|}!||!|d k}"||"z}#|#r|dz d }$| dkr|$| dz n|$}%|d d | |$f|d d | |%fz|d d | |$|fz}!t j|#d |!|}!||!m|r4t j|d }&t j|&d |d d ||f<t j| | }'t j| ft d | }(t#|D]\}}|d kr ||z })|)d k}*|*s,|)d }+||'|+|f||'|+||fz||'|+||fz},t j|(|,gd }-t j|*t j|-d |(}(|( }.|}|dkr)|.|z S|dkr|.S|dkr|.|z S|dkr|.S|.S)ab Compute TDT (Token-and-Duration Transducer) loss (https://arxiv.org/abs/2304.06795). Ported from NeMo's `TDTLossPytorch` with anti-diagonal processing. Unlike standard RNNT loss, this loss trains both the token prediction head and the duration prediction head. It uses vectorized anti-diagonal processing for efficiency: all (t, u) pairs on each anti-diagonal t+u=n are computed in parallel as batched tensor operations. Args: token_logits: Token logits of shape `(batch, T, U+1, vocab_size+1)`. duration_logits: Duration logits of shape `(batch, T, U+1, num_durations)`. targets: Target labels of shape `(batch, U)`. logit_lengths: Encoder output lengths of shape `(batch,)`. target_lengths: Target lengths of shape `(batch,)`. blank_token_id: Blank token id. durations: List of duration values (e.g., `[0, 1, 2, 3, 4]`). sigma: Logit undernormalization constant (see TDT paper). Defaults to `0.0`. reduction: Loss reduction method. One of `"mean_volume"`, `"mean_batch"`, `"mean"`, `"sum"`, or `"none"`, mirroring NeMo's `RNNTLoss` (TDT shares the same reduction knob as RNN-T). Defaults to `"mean"`, the `rnnt_reduction` of the released Parakeet TDT checkpoints. Returns: Scalar loss tensor (or per-example losses if `reduction="none"`). ) mean_volume mean_batchrsumnonezInvalid reduction mode "z". Expected one of z, c34K|]}t|VdS)N)repr).0rs X/var/www/html/CA-Chatbot/venv/lib/python3.11/site-packages/transformers/loss/loss_tdt.py ztdt_loss..>s*NqNq[\tTUwwNqNqNqNqNqNq.)dimz-inf)devicerNr)rindex)minr)maxrrrr) ValueErrorjoinr shapefloattorch log_softmaxfull unsqueezeexpandgathersqueezetensorranger%r$arange enumerateanyclampwhereappendstack logsumexprr)/rrr r r r r rrvalid_reductionsr batch_sizemax_tmax_u_token_log_probsduration_log_probs log_alphablank_log_probstargets_expandedlabel_log_probsneg_infnu_startu_end u_indices t_indicesall_candidatesidurt_prevvalid_tt_srccontribvalid_u valid_bothu_src u_src_labelstacked batch_idx log_probst_finalvalid t_clampedterminalcombinedlossess/ rtdt_lossr`sHL((( ty t tTYYNqNq`pNqNqNqEqEq t t t    F"."4Jua%%''L%++--O' "===EO*?CCC Ju5uV}}VTTTIIaaaAg&aaaAAA~&=>O qyy",,Q//66r5"EE, AAAqqq+EAI+qqq0 1",,R00    '"++  l5==888G1eema' ( ()Q)QaUQ''AE5!!L%???  M  ** / /FAs_FkG;;== LLQL''EQwwaaa 12%aaa &9:;(E9a)?@A  +g&7&7&:&:GWMM%%g... !mG 7*J~~ /"Q--!-44rps   H % %PP,P\P\P< P L P  PCyP PP \PPPPv r