o ٷi| @sddlZddlZddlZddlZddlmZddlmZddlZ ddl Z ddl m Z ddl mZddlmZddlmZddlmZdd l mZdd lmZmZmZmZdd lmZdd lmZmZdd l m!Z!ddl"m#Z#ddl$m%Z%ddl&m'Z'ddl(m)Z)ddl*m+Z+ddl,m-Z-e.e/Z0defddZ1    d*de2de2de3de3fdd Z4    d+d!e2dBd"e5e j6BdBd#e7e2dBd$e7e3dBd%e8e j9e2ff d&d'Z:Gd(d)d)ej;e!Z(sz,get_flux_post_process_func..*zvae/config.jsonblock_out_channels)vae_scale_factorimagescs |Sr) postprocess)r%image_processorrrpost_process_func6s z5get_flux_post_process_func..post_process_func) output_typemodelospathexistsrjoinopenjsonloadlenrtorchTensor)r model_name model_pathvae_config_pathf vae_configr$r)rr'rget_flux_post_process_func$s    "r;?ffffff? base_seq_len max_seq_len base_shift max_shiftcCs,||||}|||}|||}|Srr) image_seq_lenr@rArBrCmbmurrrcalculate_shift<s  rHnum_inference_stepsdevice timestepssigmasreturncKs|dur |dur td|dur>dtt|jjv}|s(td|jd|jd ||d||j}t |}||fS|durpdtt|jjv}|sZtd|jd|jd ||d ||j}t |}||fS|j|fd |i||j}||fS) a Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`. Args: scheduler (`SchedulerMixin`): The scheduler to get timesteps from. num_inference_steps (`int`): The number of diffusion steps used when generating samples with a pre-trained model. If used, `timesteps` must be `None`. device (`str` or `torch.device`, *optional*): The device to which the timesteps should be moved to. If `None`, the timesteps are not moved. timesteps (`List[int]`, *optional*): Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed, `num_inference_steps` and `sigmas` must be `None`. sigmas (`List[float]`, *optional*): Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed, `num_inference_steps` and `timesteps` must be `None`. Returns: `tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the second element is the number of inference steps. NzYOnly one of `timesteps` or `sigmas` can be passed. Please choose one to set custom valuesrKzThe current scheduler class zx's `set_timesteps` does not support custom timestep schedules. Please check whether you are using the correct scheduler.)rKrJrLzv's `set_timesteps` does not support custom sigmas schedules. Please check whether you are using the correct scheduler.)rLrJrJr) ValueErrorsetinspect signature set_timesteps parameterskeys __class__rKr3) schedulerrIrJrKrLkwargsaccepts_timesteps accept_sigmasrrrretrieve_timestepsJs2  rZc/seZdZdddedeffddZ        dVdd Z  dWd eeeBd edede j dBfddZ dXd eeeBd efddZ   dYd eeeBdeeeBd ede j dBde j dBdef ddZeddZeddZeddZ dZdd Zd!d"Zed#d$Zed%d&Zed'd(Zed)d*Zed+d,Z -d[de jde jd.e jd/e jd0e jd1e jd2e jd3e jd4e jd5ed6e jd7ed8ed9e jfd:d;Zd7edZddddd?ddd@ddAd dddddddBdCdd0gd fdDe d eeeBdBdeeeBdBdEeeeBdBdFeeeBdBd7edGedBdHedBdIedJeedBdKed edLe j!ee j!BdBd0e j dBde j dBde j dBd.e j dBd/e j dBdMedBdNedOe"ee#fdBdPeedef.dQdRZ$dSe%e&ee jfd9e'efdTdUZ(Z)S)\ FluxPipeline)prefixrr]cst||_tj|jdddddg|_t|_|j}t j |}t j |d|d|_tj |d|d|_tj |d|d|_tj |d |d|j|_t|d |_tj |d |d|_tj |d |d|_d|_t|d drvd t|jjj dnd|_!t"|d r|jdur|jj#nd|_$d|_%dS)N transformerz transformer.T) model_or_path subfolderrevisionr]fall_back_to_ptrV)r`local_files_only text_encodertext_encoder_2vae)r tokenizer tokenizer_2r!r"r#M)&super__init__rrComponentSourcer+weights_sourcesrrJr,r-r.rfrom_pretrainedrVr rdr rertorfrr^r rgr rhstagegetattrr3configr r$hasattrmodel_max_lengthtokenizer_max_lengthdefault_sample_size)selfrr]r+rcrUrrrlsF    ( zFluxPipeline.__init__Nc Cs||jddks||jddkr$td|jdd|d|d|dur7|dur7td|d|d |durJ|durJtd |d|d |durV|durVtd |durmt|tsmt|tsmtd t||durt|tst|tstd t||dur|durtd|d|d |dur|durtd|d|d |dur| durtd|dur| durtd| dur| dkrtd| dSdS)Nr!rz-`height` and `width` have to be divisible by z but are z and z(. Dimensions will be resized accordinglyzCannot forward both `prompt`: z and `prompt_embeds`: z2. Please make sure to only forward one of the two.z Cannot forward both `prompt_2`: zeProvide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined.z2`prompt` has to be of type `str` or `list` but is z4`prompt_2` has to be of type `str` or `list` but is z'Cannot forward both `negative_prompt`: z and `negative_prompt_embeds`: z)Cannot forward both `negative_prompt_2`: zIf `prompt_embeds` are provided, `pooled_prompt_embeds` also have to be passed. Make sure to generate `pooled_prompt_embeds` from the same text encoder that was used to generate `prompt_embeds`.zIf `negative_prompt_embeds` are provided, `negative_pooled_prompt_embeds` also have to be passed. Make sure to generate `negative_pooled_prompt_embeds` from the same text encoder that was used to generate `negative_prompt_embeds`.z8`max_sequence_length` cannot be greater than 512 but is )r$loggerwarningrN isinstancestrlisttype) rxpromptprompt_2heightwidthnegative_promptnegative_prompt_2 prompt_embedsnegative_prompt_embedspooled_prompt_embedsnegative_pooled_prompt_embeds"callback_on_step_end_tensor_inputsmax_sequence_lengthrrr check_inputss^$zFluxPipeline.check_inputsr"rzrnum_images_per_promptrdtypec Cst|tr|gn|}t|}t|tr|||j}|j|d|ddddd}|j}|j|dddj}|jd|jdkr]t ||s]|j |dd|d df} t d |d | |j ||jdd d } |j j}| j||jd} | j\} } } | d |d } | ||| d} | S)N max_lengthTFpt)paddingr truncation return_lengthreturn_overflowing_tokensreturn_tensorslongestrrr"zXThe following part of your input was truncated because `max_sequence_length` is set to tokens: output_hidden_statesrrrJ)r}r~r3rmaybe_convert_promptrh input_idsshaper4equal batch_decoder{r|rerprJrrepeatview) rxrrrr batch_size text_inputstext_input_idsuntruncated_ids removed_textr_seq_lenrrr_get_t5_prompt_embedss>     z"FluxPipeline._get_t5_prompt_embedsc Cst|tr|gn|}t|}t|tr|||j}|j|d|jddddd}|j}|j|dddj}|jd|jdkr`t ||s`|j |dd|jd df}t d |jd ||j||jdd }|j}|j|jj|jd }|d |}|||d}|S)NrTFr)rrrrrrrrrr"z\The following part of your input was truncated because CLIP can only handle sequences up to rrr)r}r~r3rrrgrvrrr4rrr{r|rdrprJ pooler_outputrrr) rxrrrrrrrrrrr_get_clip_prompt_embeds,s<   " z$FluxPipeline._get_clip_prompt_embedsrrrc Cst|tr|gn|}|dur+|p|}t|tr|gn|}|j||d}|j|||d}|jdur4|jjn|jj}t|j ddj |j |d}|||fS)a Args: prompt (`str` or `List[str]`, *optional*): prompt to be encoded prompt_2 (`str` or `List[str]`, *optional*): The prompt or prompts to be sent to the `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is used in all text-encoders num_images_per_prompt (`int`): number of images that should be generated per prompt prompt_embeds (`torch.FloatTensor`, *optional*): Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not provided, text embeddings will be generated from `prompt` input argument. pooled_prompt_embeds (`torch.FloatTensor`, *optional*): Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not provided, pooled text embeddings will be generated from `prompt` input argument. N)rr)rrrr"rJr) r}r~rrrdrr^r4zerosrrprJ) rxrrrrrrrtext_idsrrr encode_promptUs  zFluxPipeline.encode_promptc Cs|t||d}|dt|dddf|d<|dt|dddf|d<|j\}}}||||}|j||dS)Nr).r").r!r)r4rarangerreshaperp) rrrrJrlatent_image_idslatent_image_id_heightlatent_image_id_widthlatent_image_id_channelsrrr_prepare_latent_image_idss"" z&FluxPipeline._prepare_latent_image_idscCsR||||dd|dd}|dddddd}|||d|d|d}|S)Nr!rr"r)rpermuter)latentsrnum_channels_latentsrrrrr _pack_latentsszFluxPipeline._pack_latentscCs|j\}}}dt||d}dt||d}|||d|d|ddd}|dddddd}|||d||}|S)Nr!rrrr"r)rintrrr)rrrr$r num_patcheschannelsrrr_unpack_latentss  zFluxPipeline._unpack_latentsc Csdt||jd}dt||jd}||||f} |dur6|||d|d||} |j||d| fSt|trNt||krNtdt|d|dt| |||d}| |||||}|||d|d||} || fS)Nr!rz/You have passed a list of generators of length z+, but requested an effective batch size of z@. Make sure the batch size matches the length of the generators.) generatorrJr) rr$rrpr}rr3rNrr) rxrrrrrrJrrrrrrrprepare_latentss    zFluxPipeline.prepare_latentsc Cs||dur tdd||n|}t||jjdd|jjdd|jjdd|jjd d }t|j|||d \}}||fS) N?r"base_image_seq_lenr<max_image_seq_lenr=rBr>rCr?)rLrG)nplinspacerHrVrsgetrZ)rxrIrLrDrGrKrrrprepare_timestepss zFluxPipeline.prepare_timestepscC|jSr)_guidance_scalerxrrrguidance_scalezFluxPipeline.guidance_scalecCrr)_joint_attention_kwargsrrrrjoint_attention_kwargsrz#FluxPipeline.joint_attention_kwargscCrr)_num_timestepsrrrr num_timestepsrzFluxPipeline.num_timestepscCrr)_current_timesteprrrrcurrent_timesteprzFluxPipeline.current_timestepcCrr) _interruptrrrr interruptrzFluxPipeline.interruptFrrrrrnegative_text_idsrK do_true_cfgguidancetrue_cfg_scale cfg_normalizerMc Cs|jd| |j_t| D]N\}}|jrq||_||jdj |j |j d}||d| |||||j dd }| rI||d| |||||j dd }nd}| | | ||| }||||| }q|S)z0Diffusion loop with optional image conditioning.rriF) hidden_statestimesteprpooled_projectionsencoder_hidden_statestxt_idsimg_idsr return_dictN)rVset_begin_indexr^r enumeraterrexpandrrprJrrpredict_noise_maybe_with_cfgscheduler_step_maybe_with_cfg)rxrrrrrrrrrKrrrritrpositive_kwargsnegative_kwargs noise_predrrrdiffusesL    zFluxPipeline.diffusehas_neg_promptcCs:tdkrdS|dkrtddS|stddSdS)Nr"Tz?CFG parallel is NOT working correctly when true_cfg_scale <= 1.FzeCFG parallel is NOT working correctly when there is no negative prompt or negative prompt embeddings.)rr{r|)rxrrrrrcheck_cfg_parallel_validity9s  z(FluxPipeline.check_cfg_parallel_validityrg @pilTreqrrrrrIrLrrr*rrrc"Csdd|jDp |}tdd|jDrd}n |jr"dd|jD}|jjp+|j|j}|jjp5|j|j}|jjp;| } |jjpA| } |jj durL|jj n| } |jj pS| } |jj pY|}|jj dkrd|jj n| } |j ||||||||||||d | |_||_d|_d |_|durt|trd }n|durt|trt|}n|jd}|dup|duo|du}|d ko|}||||j||||| |d \}}}d}|r|j||||| |d \}}}|jjd }||| ||||j|j| |\}}|| | |jd \}} t||_|jj rt!j"d g| t!j#d } | $|jd} nd} |j%dur&i|_|j&||||||||||| |d d }d|_|dkrC|}!n|'||||j}||j(j)j*|j(j)j+}|j(j,|d dd}!t-|!dS)zForward pass for flux.cSs(g|]}t|tr |n|dpdqS)rr\r}r~r.0prrr e(z(FluxPipeline.forward..css(|]}t|tp|dduVqdS)rNrrrrr fs&z'FluxPipeline.forward..NcSs(g|]}t|tr dn|dpdqS)r\rrrrrrrirr)rrrrrrrrFr")rrrrrrr)r)rr)r)output).promptsallsampling_paramsrrwr$rrIrLrrrnum_outputs_per_promptrrrrrr}r~rr3rrrr^ in_channelsrrrJrrguidance_embedsr4fullfloat32rrrrrfrsscaling_factor shift_factordecoder)"rxrrrrrrrrrIrLrrrrrrrrr*rrrrrrrrrrrrKrimagerrrforwardHs                zFluxPipeline.forwardweightscCst|}||Sr)r load_weights)rxr loaderrrrr s zFluxPipeline.load_weights)NNNNNNNN)Nr"rzN)r")r"NNrzr)F)*__name__ __module__ __qualname__rr~rlrrrr4rrr FloatTensorr staticmethodrrrrrpropertyrrrrrr5boolfloatrrr Generatordictrr rtuplerOr  __classcell__rrryrr[sZ< E  0  -   1    #           E      ,)r[)r<r=r>r?)NNNN)=rPr1loggingr,collections.abcrtypingrnumpyrr4diffusers.image_processorrdiffusers.loadersr,diffusers.models.autoencoders.autoencoder_klr9diffusers.schedulers.scheduling_flow_match_euler_discreterdiffusers.utils.torch_utilsrr transformersr r r r vllm.model_executor.models.utilsrvllm_omni.diffusion.datarr,vllm_omni.diffusion.distributed.cfg_parallelr.vllm_omni.diffusion.distributed.parallel_stater%vllm_omni.diffusion.distributed.utilsr1vllm_omni.diffusion.model_loader.diffusers_loaderrvllm_omni.diffusion.models.fluxrvllm_omni.diffusion.requestr2vllm_omni.model_executor.model_loader.weight_utilsr getLoggerrr{r;rrrHr~rJrrr5rZModuler[rrrrsr                        ;