diffusion_rs_core/pipelines/flux/
mod.rs

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
use std::sync::Mutex;
use std::{cmp::Ordering, collections::HashMap, sync::Arc};

use anyhow::Result;
use diffusion_rs_common::core::{DType, Device, Tensor, D};
use diffusion_rs_common::nn::Module;
use tokenizers::Tokenizer;
use tracing::info;

use crate::models::QuantizedModel;
use crate::{
    models::{
        dispatch_load_vae_model, ClipTextConfig, ClipTextTransformer, FluxConfig, FluxModel,
        T5Config, T5EncoderModel, VAEModel,
    },
    pipelines::ComponentName,
};
use diffusion_rs_common::{from_mmaped_safetensors, ModelSource};

use super::sampling::Sampler;
use super::scheduler::SchedulerConfig;
use super::{ComponentElem, DiffusionGenerationParams, Loader, ModelPipeline, Offloading};

mod sampling;

pub struct FluxLoader;

impl Loader for FluxLoader {
    fn name(&self) -> &'static str {
        "flux"
    }

    fn required_component_names(&self) -> Vec<ComponentName> {
        vec![
            ComponentName::Scheduler,
            ComponentName::TextEncoder(1),
            ComponentName::TextEncoder(2),
            ComponentName::Tokenizer(1),
            ComponentName::Tokenizer(2),
            ComponentName::Transformer,
            ComponentName::Vae,
        ]
    }

    fn load_from_components(
        &self,
        mut components: HashMap<ComponentName, ComponentElem>,
        device: &Device,
        dtype: DType,
        silent: bool,
        offloading_type: Option<Offloading>,
        source: Arc<ModelSource>,
    ) -> Result<Arc<Mutex<dyn ModelPipeline>>> {
        let scheduler = components.remove(&ComponentName::Scheduler).unwrap();
        let clip_component = components.remove(&ComponentName::TextEncoder(1)).unwrap();
        let t5_component = components.remove(&ComponentName::TextEncoder(2)).unwrap();
        let clip_tok_component = components.remove(&ComponentName::Tokenizer(1)).unwrap();
        let t5_tok_component = components.remove(&ComponentName::Tokenizer(2)).unwrap();
        let flux_component = components.remove(&ComponentName::Transformer).unwrap();
        let vae_component = components.remove(&ComponentName::Vae).unwrap();

        let t5_flux_device = match offloading_type {
            Some(Offloading::Full) => Device::Cpu,
            None => device.clone(),
        };

        let scheduler_config = if let ComponentElem::Config { files } = scheduler {
            serde_json::from_str::<SchedulerConfig>(
                &files["scheduler/scheduler_config.json"].read_to_string(&source)?,
            )?
        } else {
            anyhow::bail!("expected scheduler config")
        };
        let clip_tokenizer = if let ComponentElem::Other { files } = clip_tok_component {
            let vocab_file = &files["tokenizer/vocab.json"];
            let merges_file = &files["tokenizer/merges.txt"];

            diffusion_rs_common::load_bpe_tokenizer(vocab_file, merges_file, &source)?
        } else {
            anyhow::bail!("incorrect storage of clip tokenizer")
        };
        let t5_tokenizer = if let ComponentElem::Other { files } = t5_tok_component {
            Tokenizer::from_bytes(files["tokenizer_2/tokenizer.json"].read_to_string(&source)?)
                .map_err(anyhow::Error::msg)?
        } else {
            anyhow::bail!("incorrect storage of t5 tokenizer")
        };
        if !silent {
            info!("loading CLIP model");
        }
        let clip_component = if let ComponentElem::Model {
            safetensors,
            config,
        } = clip_component
        {
            let cfg: ClipTextConfig = serde_json::from_str(&config.read_to_string(&source)?)?;

            let vb = from_mmaped_safetensors(
                safetensors.into_values().collect(),
                Some(dtype),
                device,
                silent,
                source.clone(),
            )?;
            ClipTextTransformer::new(vb.pp("text_model"), &cfg)?
        } else {
            anyhow::bail!("incorrect storage of clip model")
        };
        if !silent {
            info!("loading T5 model");
        }
        let t5_component = if let ComponentElem::Model {
            safetensors,
            config,
        } = t5_component
        {
            let cfg: T5Config = serde_json::from_str(&config.read_to_string(&source)?)?;
            let vb = from_mmaped_safetensors(
                safetensors.into_values().collect(),
                Some(dtype),
                &t5_flux_device,
                silent,
                source.clone(),
            )?;
            T5EncoderModel::new(vb, &cfg)?
        } else {
            anyhow::bail!("incorrect storage of t5 model")
        };
        if !silent {
            info!("loading VAE model");
        }
        let vae_component = if let ComponentElem::Model {
            safetensors,
            config,
        } = vae_component
        {
            dispatch_load_vae_model(
                &config,
                safetensors.into_values().collect(),
                device,
                dtype,
                silent,
                source.clone(),
            )?
        } else {
            anyhow::bail!("incorrect storage of vae model")
        };
        if !silent {
            info!("loading FLUX model");
        }
        let flux_component = if let ComponentElem::Model {
            safetensors,
            config,
        } = flux_component
        {
            let cfg: FluxConfig = serde_json::from_str(&config.read_to_string(&source)?)?;
            let vb = from_mmaped_safetensors(
                safetensors.into_values().collect(),
                Some(dtype),
                &t5_flux_device,
                silent,
                source,
            )?;
            FluxModel::new(&cfg, vb)?
        } else {
            anyhow::bail!("incorrect storage of flux model")
        };

        if !silent {
            info!(
                "FLUX pipeline using a guidance-distilled model: {}",
                flux_component.is_guidance()
            );
        }

        let pipeline = FluxPipeline {
            clip_tokenizer: Arc::new(clip_tokenizer),
            clip_model: clip_component,
            t5_tokenizer: Arc::new(t5_tokenizer),
            t5_model: t5_component,
            vae_model: vae_component,
            flux_model: flux_component,
            scheduler_config,
            device: device.clone(),
        };

        Ok(Arc::new(Mutex::new(pipeline)))
    }
}

pub struct FluxPipeline {
    clip_tokenizer: Arc<Tokenizer>,
    clip_model: ClipTextTransformer,
    t5_tokenizer: Arc<Tokenizer>,
    t5_model: T5EncoderModel,
    vae_model: Arc<dyn VAEModel>,
    flux_model: FluxModel,
    scheduler_config: SchedulerConfig,
    device: Device,
}

impl FluxPipeline {
    fn tokenize_and_pad(
        prompts: Vec<String>,
        tokenizer: &Tokenizer,
    ) -> diffusion_rs_common::core::Result<Vec<Vec<u32>>> {
        let mut t5_tokens = Vec::new();
        let unpadded_t5_tokens = tokenizer
            .encode_batch(prompts, true)
            .map_err(|e| diffusion_rs_common::core::Error::Msg(e.to_string()))?
            .into_iter()
            .map(|e| e.get_ids().to_vec())
            .collect::<Vec<_>>();
        let t5_max_tokens = unpadded_t5_tokens.iter().map(|x| x.len()).max().unwrap();
        for mut tokenization in unpadded_t5_tokens {
            tokenization.extend(vec![0; t5_max_tokens - tokenization.len()]);
            t5_tokens.push(tokenization);
        }

        Ok(t5_tokens)
    }
}

impl ModelPipeline for FluxPipeline {
    fn forward(
        &mut self,
        prompts: Vec<String>,
        params: DiffusionGenerationParams,
        offloading_type: Option<Offloading>,
    ) -> diffusion_rs_common::core::Result<Tensor> {
        match offloading_type {
            Some(Offloading::Full) => {
                self.t5_model.to_device(&self.device)?;
            }
            None => (),
        }

        let mut t5_input_ids = Tensor::new(
            Self::tokenize_and_pad(prompts.clone(), &self.t5_tokenizer)?,
            &self.device,
        )?;

        if !self.flux_model.is_guidance() {
            match t5_input_ids.dim(1)?.cmp(&256) {
                Ordering::Greater => {
                    diffusion_rs_common::bail!("T5 embedding length greater than 256, please shrink the prompt or use the -dev (with guidance distillation) version.")
                }
                Ordering::Less | Ordering::Equal => {
                    t5_input_ids =
                        t5_input_ids.pad_with_zeros(D::Minus1, 0, 256 - t5_input_ids.dim(1)?)?;
                }
            }
        }

        let t5_embed = self.t5_model.forward(&t5_input_ids)?;

        match offloading_type {
            Some(Offloading::Full) => {
                self.t5_model.to_device(&Device::Cpu)?;
            }
            None => (),
        }

        let clip_input_ids = Tensor::new(
            Self::tokenize_and_pad(prompts, &self.clip_tokenizer)?,
            self.clip_model.device(),
        )?;
        let clip_embed = self.clip_model.forward(&clip_input_ids)?;

        let mut img = sampling::get_noise(
            t5_embed.dim(0)?,
            params.height,
            params.width,
            t5_embed.device(),
        )?
        .to_dtype(t5_embed.dtype())?;

        let state = sampling::State::new(&t5_embed, &clip_embed, &img)?;
        let mu = sampling::calculate_shift(
            img.dims()[1],
            self.scheduler_config.base_image_seq_len,
            self.scheduler_config.max_image_seq_len,
            self.scheduler_config.base_shift,
            self.scheduler_config.max_shift,
        );
        let timesteps = self
            .scheduler_config
            .get_timesteps(params.num_steps, Some(mu))?;

        let bs = img.dim(0)?;
        let dev = img.device();

        match offloading_type {
            Some(Offloading::Full) => {
                self.flux_model.to_device(&self.device)?;
            }
            None => (),
        }

        let guidance = if self.flux_model.is_guidance() {
            Some(Tensor::full(params.guidance_scale as f32, bs, dev)?)
        } else {
            None
        };
        let step = |img: &Tensor, t_vec: &Tensor| -> diffusion_rs_common::core::Result<Tensor> {
            self.flux_model.forward(
                img,
                &state.img_ids,
                &state.txt,
                &state.txt_ids,
                t_vec,
                &state.vec,
                guidance.as_ref(),
            )
        };

        let sampler = Sampler::new(&self.scheduler_config.scheduler_type);
        img = sampler.sample(&timesteps, &state.img, step)?;

        match offloading_type {
            Some(Offloading::Full) => {
                self.flux_model.to_device(&Device::Cpu)?;
            }
            None => (),
        }

        img = sampling::unpack(&img, params.height, params.width)?;

        img = ((img / self.vae_model.scale_factor())? + self.vae_model.shift_factor())?;
        img = self.vae_model.decode(&img)?;

        img = ((img.clamp(-1f32, 1f32)? + 1.0)? * 127.5)?.to_dtype(DType::U8)?;

        Ok(img)
    }
}