mistralrs_core/vision_models/llava/llava_llm/
mistral.rs

1#![allow(clippy::cast_possible_truncation, clippy::cast_precision_loss)]
2
3use std::sync::Arc;
4
5/// Mistral LLM, https://github.com/mistralai/mistral-src
6use candle_core::{DType, Device, Module, Result, Tensor};
7use mistralrs_quant::{
8    ColumnParallelLayer, QuantMethod, QuantMethodConfig, RowParallelLayer, ShardedVarBuilder,
9    UnquantLinear,
10};
11
12use crate::{
13    amoe::{AnyMoeBaseModelMixin, AnyMoeTrainableLayer, MlpLayer, MoeMlp},
14    attention::SdpaParams,
15    device_map::DeviceMapper,
16    get_delta_from_lora_ab,
17    layers::{self, linear_no_bias, Activation, CausalMasker, MatMul, RmsNorm, Sdpa},
18    layers_masker::PastKvLenCache,
19    paged_attention::{AttentionImplementation, ModelConfigMetadata, PagedAttention},
20    pipeline::{
21        extract_logits,
22        text_models_inputs_processor::{FlashParams, PagedAttentionInputMetadata},
23        Cache, EitherCache, IsqModel, NormalLoadingMetadata, NormalModel,
24    },
25    utils::progress::NiceProgressBar,
26    AnyMoeConfig, AnyMoeExpertType,
27};
28
29use super::{LLaVALLM, OrdinaryRoPE};
30use crate::models::mistral::Config;
31
32#[derive(Clone)]
33#[allow(clippy::upper_case_acronyms)]
34struct MLP {
35    gate_proj: Arc<dyn QuantMethod>,
36    up_proj: Arc<dyn QuantMethod>,
37    down_proj: Arc<dyn QuantMethod>,
38    act_fn: Activation,
39    params: Vec<usize>,
40}
41
42impl MLP {
43    fn new(cfg: &Config, vb: ShardedVarBuilder, comm: &Arc<mistralrs_quant::Comm>) -> Result<Self> {
44        let hidden_sz = cfg.hidden_size;
45        let intermediate_sz = cfg.intermediate_size;
46        let gate_proj = ColumnParallelLayer::new(
47            hidden_sz,
48            intermediate_sz,
49            &cfg.quantization_config,
50            false,
51            comm,
52            vb.pp("gate_proj"),
53        )?;
54        let up_proj = ColumnParallelLayer::new(
55            hidden_sz,
56            intermediate_sz,
57            &cfg.quantization_config,
58            false,
59            comm,
60            vb.pp("up_proj"),
61        )?;
62        let down_proj = RowParallelLayer::new(
63            intermediate_sz,
64            hidden_sz,
65            &cfg.quantization_config,
66            false,
67            comm,
68            vb.pp("down_proj"),
69        )?;
70        Ok(Self {
71            gate_proj,
72            up_proj,
73            down_proj,
74            act_fn: cfg.hidden_act,
75            params: vec![hidden_sz, intermediate_sz],
76        })
77    }
78}
79
80impl AnyMoeTrainableLayer for MLP {}
81
82impl MlpLayer for MLP {
83    fn forward(&self, xs: &Tensor) -> Result<Tensor> {
84        let original_dtype = xs.dtype();
85        let mut xs = xs.clone();
86        if let Some(t) = self.gate_proj.quantized_act_type() {
87            xs = xs.to_dtype(t)?;
88        }
89        let lhs = MatMul
90            .qmethod_matmul(&xs, &*self.gate_proj)?
91            .apply(&self.act_fn)?;
92        let rhs = MatMul.qmethod_matmul(&xs, &*self.up_proj)?;
93        let mut res = MatMul.qmethod_matmul(&(lhs * rhs)?, &*self.down_proj)?;
94        if self.gate_proj.quantized_act_type().is_some() {
95            res = res.to_dtype(original_dtype)?;
96        }
97        Ok(res)
98    }
99    fn get_isq_layers(&mut self) -> Vec<&mut Arc<dyn QuantMethod>> {
100        vec![&mut self.gate_proj, &mut self.up_proj, &mut self.down_proj]
101    }
102    fn clone(&self) -> Box<dyn MlpLayer> {
103        Box::new(Clone::clone(self))
104    }
105    fn get_params(&self) -> &[usize] {
106        &self.params
107    }
108    fn hidden_act(&self) -> Activation {
109        self.act_fn
110    }
111    // gate, up, down
112    fn new_added_delta(&self, deltas: Vec<Option<Tensor>>) -> Result<Box<dyn MlpLayer>> {
113        let gate_proj = if let Some(ref delta) = deltas[0] {
114            self.gate_proj.add_delta_w(delta)?
115        } else {
116            self.gate_proj.clone()
117        };
118        let up_proj = if let Some(ref delta) = deltas[1] {
119            self.up_proj.add_delta_w(delta)?
120        } else {
121            self.up_proj.clone()
122        };
123        let down_proj = if let Some(ref delta) = deltas[2] {
124            self.down_proj.add_delta_w(delta)?
125        } else {
126            self.down_proj.clone()
127        };
128
129        Ok(Box::new(Self {
130            gate_proj,
131            up_proj,
132            down_proj,
133            act_fn: self.act_fn,
134            params: self.params.clone(),
135        }))
136    }
137
138    fn dtype_device(&self) -> (DType, Device) {
139        self.gate_proj.dtype_and_device()
140    }
141}
142
143struct Attention {
144    q_proj: Arc<dyn QuantMethod>,
145    k_proj: Arc<dyn QuantMethod>,
146    v_proj: Arc<dyn QuantMethod>,
147    o_proj: Arc<dyn QuantMethod>,
148    num_heads: usize,
149    num_kv_heads: usize,
150    head_dim: usize,
151    sliding_window: Option<usize>,
152    paged_attn: Option<PagedAttention>,
153    sdpa_params: SdpaParams,
154}
155
156impl Attention {
157    fn new(
158        cfg: &Config,
159        vb: ShardedVarBuilder,
160        paged_attn: Option<PagedAttention>,
161        comm: &Arc<mistralrs_quant::Comm>,
162    ) -> Result<Self> {
163        let hidden_sz = cfg.hidden_size;
164        let num_heads = cfg.num_attention_heads;
165        let num_kv_heads = cfg.num_key_value_heads;
166        let head_dim = cfg.head_dim();
167        let q_proj = ColumnParallelLayer::new(
168            hidden_sz,
169            num_heads * head_dim,
170            &cfg.quantization_config,
171            false,
172            comm,
173            vb.pp("q_proj"),
174        )?;
175        let kv_shard = mistralrs_quant::compute_kv_shard(
176            cfg.num_key_value_heads,
177            cfg.hidden_size / cfg.num_attention_heads,
178            comm,
179        );
180        let k_proj = ColumnParallelLayer::new_with_shard(
181            hidden_sz,
182            num_kv_heads * head_dim,
183            &cfg.quantization_config,
184            false,
185            comm,
186            kv_shard,
187            vb.pp("k_proj"),
188        )?;
189        let v_proj = ColumnParallelLayer::new_with_shard(
190            hidden_sz,
191            num_kv_heads * head_dim,
192            &cfg.quantization_config,
193            false,
194            comm,
195            kv_shard,
196            vb.pp("v_proj"),
197        )?;
198        let o_proj = RowParallelLayer::new(
199            num_heads * head_dim,
200            hidden_sz,
201            &cfg.quantization_config,
202            false,
203            comm,
204            vb.pp("o_proj"),
205        )?;
206        Ok(Self {
207            q_proj,
208            k_proj,
209            v_proj,
210            o_proj,
211            num_heads: num_heads / comm.world_size(),
212            num_kv_heads: (num_kv_heads / comm.world_size()).max(1),
213            head_dim,
214            sliding_window: cfg.sliding_window,
215            paged_attn,
216            sdpa_params: SdpaParams {
217                n_kv_groups: mistralrs_quant::compute_n_kv_groups(
218                    cfg.num_key_value_heads,
219                    cfg.num_attention_heads,
220                    comm,
221                ),
222                softcap: None,
223                softmax_scale: 1.0 / (head_dim as f32).sqrt(),
224                sliding_window: cfg.sliding_window,
225            },
226        })
227    }
228
229    #[allow(clippy::too_many_arguments)]
230    fn forward(
231        &self,
232        xs: &Tensor,
233        attention_mask: Option<&Tensor>,
234        seqlen_offsets: &[usize],
235        kv_cache: &mut Option<(Tensor, Tensor)>,
236        rope_parameter: (&Tensor, &Tensor),
237        metadata: Option<((Tensor, Tensor), &PagedAttentionInputMetadata)>,
238        flash_params: &FlashParams,
239    ) -> Result<Tensor> {
240        let (b_sz, q_len, _) = xs.dims3()?;
241
242        let original_dtype = xs.dtype();
243        let mut xs = xs.clone();
244        if let Some(t) = self.q_proj.quantized_act_type() {
245            xs = xs.to_dtype(t)?;
246        }
247        let mut q = MatMul.qmethod_matmul(&xs, &*self.q_proj)?;
248        let mut k = MatMul.qmethod_matmul(&xs, &*self.k_proj)?;
249        let mut v = MatMul.qmethod_matmul(&xs, &*self.v_proj)?;
250        if self.q_proj.quantized_act_type().is_some() {
251            q = q.to_dtype(original_dtype)?;
252            k = k.to_dtype(original_dtype)?;
253            v = v.to_dtype(original_dtype)?;
254        }
255
256        let mut q = q
257            .reshape((b_sz, q_len, self.num_heads, self.head_dim))?
258            .transpose(1, 2)?
259            .contiguous()?;
260        let mut k = k
261            .reshape((b_sz, q_len, self.num_kv_heads, self.head_dim))?
262            .transpose(1, 2)?
263            .contiguous()?;
264        q = OrdinaryRoPE::forward(&q, seqlen_offsets[0], rope_parameter.0, rope_parameter.1)?;
265        k = OrdinaryRoPE::forward(&k, seqlen_offsets[0], rope_parameter.0, rope_parameter.1)?;
266        let v = v
267            .reshape((b_sz, q_len, self.num_kv_heads, self.head_dim))?
268            .transpose(1, 2)?;
269
270        let mut attn_output = match &self.paged_attn {
271            Some(paged_attn) => match metadata {
272                Some(((key_cache, value_cache), input_metadata)) => paged_attn.forward(
273                    &q,
274                    &k,
275                    &v,
276                    attention_mask,
277                    Some(key_cache),
278                    Some(value_cache),
279                    input_metadata,
280                    &self.sdpa_params,
281                    Some(flash_params),
282                )?,
283                None => {
284                    // If we don't have metadata, we are most likely generating an imatrix so we don't want to populate that.
285                    // Generating the dummy metadata with the assumption that we are not generating text (only processing prompts).
286                    let input_metadata = PagedAttentionInputMetadata::dummy(q.device())?;
287                    // Sanity check.
288                    assert!(attention_mask.is_some());
289                    paged_attn.forward(
290                        &q,
291                        &k,
292                        &v,
293                        attention_mask,
294                        None,
295                        None,
296                        &input_metadata,
297                        &self.sdpa_params,
298                        Some(flash_params),
299                    )?
300                }
301            },
302            None => {
303                let (k, v, attn_mask) = Cache::update_kv_cache_sliding_window(
304                    kv_cache,
305                    k,
306                    v,
307                    attention_mask,
308                    self.sliding_window,
309                    false,
310                )?;
311
312                Sdpa.run_attention(
313                    &q,
314                    &k,
315                    &v,
316                    attn_mask.as_ref(),
317                    Some(flash_params),
318                    &self.sdpa_params,
319                )?
320            }
321        };
322
323        if let Some(t) = self.q_proj.quantized_act_type() {
324            attn_output = attn_output.to_dtype(t)?;
325        }
326        attn_output = if attention_mask.is_some() {
327            attn_output.transpose(1, 2)?.reshape((b_sz, q_len, ()))?
328        } else {
329            attn_output.reshape((b_sz, q_len, ()))?
330        };
331        let mut res = MatMul.qmethod_matmul(&attn_output, &*self.o_proj)?;
332        if self.q_proj.quantized_act_type().is_some() {
333            res = res.to_dtype(original_dtype)?;
334        }
335        Ok(res)
336    }
337}
338
339struct DecoderLayer {
340    self_attn: Attention,
341    mlp: Box<dyn MlpLayer>,
342    input_layernorm: RmsNorm,
343    post_attention_layernorm: RmsNorm,
344    rope_parameter: (Tensor, Tensor),
345}
346
347impl DecoderLayer {
348    #[allow(clippy::too_many_arguments)]
349    fn new(
350        cfg: &Config,
351        vb: ShardedVarBuilder,
352        mapper: &dyn DeviceMapper,
353        layer_idx: usize,
354        loading_isq: bool,
355        paged_attn: Option<PagedAttention>,
356        rope_parameter: (Tensor, Tensor),
357        comm: &Arc<mistralrs_quant::Comm>,
358    ) -> Result<Self> {
359        let self_attn = Attention::new(
360            cfg,
361            mapper.set_device(layer_idx, vb.pp("self_attn"), loading_isq),
362            paged_attn,
363            comm,
364        )?;
365        let mlp = MLP::new(
366            cfg,
367            mapper.set_device(layer_idx, vb.pp("mlp"), loading_isq),
368            comm,
369        )?;
370        let input_layernorm = RmsNorm::new(
371            cfg.hidden_size,
372            cfg.rms_norm_eps,
373            mapper.set_device(layer_idx, vb.pp("input_layernorm"), false),
374        )?;
375        let post_attention_layernorm = RmsNorm::new(
376            cfg.hidden_size,
377            cfg.rms_norm_eps,
378            mapper.set_device(layer_idx, vb.pp("post_attention_layernorm"), false),
379        )?;
380        Ok(Self {
381            self_attn,
382            mlp: Box::new(mlp),
383            input_layernorm,
384            post_attention_layernorm,
385            rope_parameter,
386        })
387    }
388
389    #[allow(clippy::too_many_arguments)]
390    fn forward(
391        &self,
392        xs: &Tensor,
393        attention_mask: Option<&Tensor>,
394        seqlen_offsets: &[usize],
395        kv_cache: &mut Option<(Tensor, Tensor)>,
396        metadata: Option<((Tensor, Tensor), &PagedAttentionInputMetadata)>,
397        flash_params: &FlashParams,
398    ) -> Result<Tensor> {
399        let residual = xs;
400        let mut xs = self.input_layernorm.forward(xs)?;
401        xs = self.self_attn.forward(
402            &xs,
403            attention_mask,
404            seqlen_offsets,
405            kv_cache,
406            (&self.rope_parameter.0, &self.rope_parameter.1),
407            metadata,
408            flash_params,
409        )?;
410        xs = (xs + residual)?;
411        let residual = &xs;
412        let xs = self
413            .mlp
414            .forward(&xs.apply(&self.post_attention_layernorm)?)?;
415        residual + xs
416    }
417}
418
419pub struct Model {
420    embed_tokens: candle_nn::Embedding,
421    layers: Vec<DecoderLayer>,
422    norm: RmsNorm,
423    lm_head: Arc<dyn QuantMethod>,
424    sliding_window: Option<usize>,
425    device: Device,
426    cache: EitherCache,
427    max_seq_len: usize,
428    mapper: Box<dyn DeviceMapper + Send + Sync>,
429    cfg: ModelConfigMetadata,
430}
431
432impl Model {
433    pub fn new(
434        cfg: &Config,
435        vb: ShardedVarBuilder,
436        is_gptx: bool,
437        normal_loading_metadata: NormalLoadingMetadata,
438        attention_mechanism: AttentionImplementation,
439    ) -> Result<Self> {
440        let vb_m = vb.pp("model");
441        let vb_lm_head = vb.pp("lm_head");
442        Self::new_inner(
443            cfg,
444            vb_m,
445            vb_lm_head,
446            is_gptx,
447            normal_loading_metadata,
448            attention_mechanism,
449        )
450    }
451
452    pub fn new_inner(
453        cfg: &Config,
454        vb_m: ShardedVarBuilder,
455        vb_lm_head: ShardedVarBuilder,
456        _is_gptx: bool,
457        normal_loading_metadata: NormalLoadingMetadata,
458        attention_mechanism: AttentionImplementation,
459    ) -> Result<Self> {
460        if let Some(ref quant_cfg) = &cfg.quantization_config {
461            tracing::info!(
462                "Using {} quantization: {}.",
463                quant_cfg.name(),
464                quant_cfg.get_bits_name(&vb_m)
465            );
466        }
467        let mapper = normal_loading_metadata.mapper;
468        let embed_tokens = layers::embedding(
469            cfg.vocab_size,
470            cfg.hidden_size,
471            mapper.set_nm_device(vb_m.pp("embed_tokens"), false),
472            &cfg.quantization_config,
473        )?;
474        let head_dim = cfg.hidden_size / cfg.num_attention_heads;
475        let vb_l = vb_m.pp("layers");
476        let layers = NiceProgressBar::<_, 'b'>(
477            0..cfg.num_hidden_layers,
478            "Loading repeating layers",
479            &normal_loading_metadata.multi_progress,
480        )
481        .par_iter_if_isq(|layer_idx| {
482            let device = mapper
483                .device_for(layer_idx, false)
484                .unwrap_or(&normal_loading_metadata.real_device);
485            let rope_parameters = OrdinaryRoPE::create_parameters(
486                head_dim,
487                cfg.max_position_embeddings,
488                cfg.rope_theta as f32,
489                vb_m.dtype(),
490                device,
491            )?;
492            let paged_attn = match &attention_mechanism {
493                AttentionImplementation::Eager => None,
494                AttentionImplementation::PagedAttention => {
495                    Some(PagedAttention::new(head_dim, device, None)?)
496                }
497            };
498            let comm = mapper.get_comm_for(layer_idx)?;
499            DecoderLayer::new(
500                cfg,
501                vb_l.pp(layer_idx),
502                &*mapper,
503                layer_idx,
504                normal_loading_metadata.loading_isq,
505                paged_attn,
506                rope_parameters,
507                &comm,
508            )
509        })?;
510        let norm = RmsNorm::new(
511            cfg.hidden_size,
512            cfg.rms_norm_eps,
513            mapper.set_nm_device(vb_m.pp("norm"), false),
514        )?;
515        let lm_head = linear_no_bias(
516            cfg.hidden_size,
517            cfg.vocab_size,
518            mapper.set_nm_device(vb_lm_head, normal_loading_metadata.loading_isq),
519        )?;
520        Ok(Self {
521            embed_tokens,
522            layers,
523            norm,
524            lm_head: Arc::new(UnquantLinear::new(QuantMethodConfig::Unquantized(lm_head))?),
525            sliding_window: cfg.sliding_window,
526            device: normal_loading_metadata.real_device,
527            cache: EitherCache::Full(Cache::new(cfg.num_hidden_layers, false)),
528            max_seq_len: cfg.max_position_embeddings,
529            cfg: ModelConfigMetadata {
530                max_seq_len: cfg.max_position_embeddings,
531                num_layers: cfg.num_hidden_layers,
532                hidden_size: cfg.hidden_size,
533                num_attn_heads: cfg.num_attention_heads / mapper.get_comm_for(0)?.world_size(),
534                num_kv_heads: (cfg.num_key_value_heads / mapper.get_comm_for(0)?.world_size())
535                    .max(1),
536                sliding_window: cfg.sliding_window,
537                k_head_dim: cfg.head_dim(),
538                v_head_dim: cfg.head_dim(),
539            },
540            mapper,
541        })
542    }
543
544    pub fn get_input_embeddings(&self, input_ids: &Tensor) -> Result<Tensor> {
545        self.embed_tokens.forward(input_ids)
546    }
547
548    pub fn forward(
549        &self,
550        input_ids: &Tensor,
551        seqlen_offsets: &[usize],
552        context_lens: Vec<(usize, usize)>,
553        metadata: Option<(Vec<(Tensor, Tensor)>, &PagedAttentionInputMetadata)>,
554        flash_params: &FlashParams,
555    ) -> Result<Tensor> {
556        self.forward_embeds(
557            input_ids,
558            self.embed_tokens.forward(input_ids)?,
559            seqlen_offsets,
560            context_lens,
561            metadata,
562            flash_params,
563        )
564    }
565
566    #[allow(clippy::too_many_arguments)]
567    pub fn forward_embeds(
568        &self,
569        input_ids: &Tensor,
570        input_embeds: Tensor,
571        seqlen_offsets: &[usize],
572        context_lens: Vec<(usize, usize)>,
573        metadata: Option<(Vec<(Tensor, Tensor)>, &PagedAttentionInputMetadata)>,
574        flash_params: &FlashParams,
575    ) -> Result<Tensor> {
576        let mut xs = input_embeds;
577        let mut cache = self.cache.full().lock();
578        let attention_mask = CausalMasker.make_sliding_window_causal_mask_matrix(
579            input_ids,
580            metadata
581                .as_ref()
582                .map(|(_, _)| &seqlen_offsets as &dyn PastKvLenCache)
583                .unwrap_or(&*cache as &dyn PastKvLenCache),
584            self.sliding_window,
585            xs.dtype(),
586            self.cfg.num_attn_heads,
587        )?;
588        let attention_mask = attention_mask.filter(|_| {
589            metadata
590                .as_ref()
591                .map(|(_, meta)| meta.is_first_prompt_chunk)
592                .unwrap_or(true)
593        });
594        for (i, layer) in self.layers.iter().enumerate() {
595            xs = self.mapper.map(xs, i)?;
596            xs = layer.forward(
597                &xs,
598                attention_mask
599                    .as_ref()
600                    .map(|m| m.to_device(xs.device()).unwrap())
601                    .as_ref(),
602                seqlen_offsets,
603                &mut cache[i],
604                metadata
605                    .as_ref()
606                    .map(|(kv_cache, metadata)| (kv_cache[i].clone(), *metadata)),
607                flash_params,
608            )?;
609        }
610        xs = xs.to_device(&self.device)?;
611        xs = xs.apply(&self.norm)?;
612        if let Some(t) = self.lm_head.quantized_act_type() {
613            xs = xs.to_dtype(t)?;
614        }
615        extract_logits(&MatMul.qmethod_matmul(&xs, &*self.lm_head)?, context_lens)
616    }
617}
618
619impl IsqModel for Model {
620    fn get_layers(
621        &mut self,
622    ) -> (
623        Vec<(&mut Arc<dyn QuantMethod>, Option<usize>)>,
624        &dyn DeviceMapper,
625    ) {
626        let mut tensors = Vec::new();
627        tensors.push((&mut self.lm_head, None));
628        for (i, layer) in self.layers.iter_mut().enumerate() {
629            tensors.push((&mut layer.self_attn.q_proj, Some(i)));
630            tensors.push((&mut layer.self_attn.k_proj, Some(i)));
631            tensors.push((&mut layer.self_attn.v_proj, Some(i)));
632            tensors.push((&mut layer.self_attn.o_proj, Some(i)));
633            tensors.extend(
634                layer
635                    .mlp
636                    .get_isq_layers()
637                    .into_iter()
638                    .map(|m| (m, Some(i)))
639                    .collect::<Vec<_>>(),
640            );
641        }
642        (tensors, &*self.mapper)
643    }
644
645    fn residual_tensors(&self) -> Vec<(String, Tensor)> {
646        Vec::new()
647    }
648}
649
650impl LLaVALLM for Model {
651    fn embed(&self, input_ids: &Tensor) -> Result<Tensor> {
652        self.get_input_embeddings(input_ids)
653    }
654
655    fn forward_input_embed(
656        &self,
657        input_ids: &Tensor,
658        input_embed: Tensor,
659        seqlen_offsets: &[usize],
660        context_lens: Vec<(usize, usize)>,
661        metadata: Option<(Vec<(Tensor, Tensor)>, &PagedAttentionInputMetadata)>,
662        flash_params: &FlashParams,
663    ) -> Result<Tensor> {
664        self.forward_embeds(
665            input_ids,
666            input_embed,
667            seqlen_offsets,
668            context_lens,
669            metadata,
670            flash_params,
671        )
672    }
673}
674
675impl NormalModel for Model {
676    fn forward(
677        &self,
678        input_ids: &Tensor,
679        seqlen_offsets: &[usize],
680        context_lens: Vec<(usize, usize)>,
681        _position_ids: Vec<usize>,
682        metadata: Option<(Vec<(Tensor, Tensor)>, &PagedAttentionInputMetadata)>,
683        flash_params: &FlashParams,
684    ) -> Result<Tensor> {
685        self.forward(
686            input_ids,
687            seqlen_offsets,
688            context_lens,
689            metadata,
690            flash_params,
691        )
692    }
693    fn xlora_forward(
694        &self,
695        _input_ids: &Tensor,
696        _input_ids_full: &Tensor,
697        _seqlen_offsets: &[usize],
698        _seqlen_offsets_full: &[usize],
699        _no_kv_cache: bool,
700        _non_granular_state: &Option<crate::xlora_models::NonGranularState>,
701        _context_lens: Vec<(usize, usize)>,
702        _position_ids: Vec<usize>,
703        _flash_params: &FlashParams,
704        _flash_params_full: &FlashParams,
705    ) -> Result<Tensor> {
706        unimplemented!()
707    }
708    fn cache(&self) -> &EitherCache {
709        &self.cache
710    }
711    fn cache_mut(&mut self) -> &mut EitherCache {
712        &mut self.cache
713    }
714    fn device(&self) -> &Device {
715        &self.device
716    }
717    fn is_xlora(&self) -> bool {
718        false
719    }
720    fn max_seq_len(&self) -> usize {
721        self.max_seq_len
722    }
723    fn config(&self) -> &ModelConfigMetadata {
724        &self.cfg
725    }
726}
727
728impl AnyMoeBaseModelMixin for Model {
729    fn get_mlps(&self) -> Vec<&dyn MlpLayer> {
730        let mut mlps = Vec::new();
731        for layer in &self.layers {
732            mlps.push(&*layer.mlp);
733        }
734        mlps
735    }
736    fn get_mlps_mut(&mut self) -> Vec<&mut Box<dyn MlpLayer>> {
737        let mut mlps = Vec::new();
738        for layer in &mut self.layers {
739            mlps.push(&mut layer.mlp);
740        }
741        mlps
742    }
743    fn create_anymoe_layers(
744        &mut self,
745        additional_vbs: Vec<ShardedVarBuilder>,
746        config: AnyMoeConfig,
747        (prefix, mlp): (String, String),
748        mut layers: Vec<usize>,
749        expert_type: AnyMoeExpertType,
750        gate_vb: Option<ShardedVarBuilder>,
751    ) -> Result<()> {
752        let mut experts: Vec<Vec<Box<dyn MlpLayer>>> = Vec::new();
753        if layers.is_empty() {
754            layers = (0..self.layers.len()).collect::<Vec<_>>();
755        }
756        for _ in 0..layers.len() {
757            experts.push(Vec::new());
758        }
759        for vb in additional_vbs {
760            let vb = vb.pp(&prefix);
761            for (layer, row) in experts.iter_mut().enumerate() {
762                if !layers.contains(&layer) {
763                    continue;
764                }
765
766                let intermediate_size = self.layers[layer].mlp.get_params()[1];
767                let hidden_size = self.layers[layer].mlp.get_params()[0];
768                match expert_type {
769                    AnyMoeExpertType::FineTuned => {
770                        let (dtype, device) = self.layers[layer].mlp.dtype_device();
771                        row.push(Box::new(MLP::new(
772                            &Config {
773                                intermediate_size: self.layers[layer].mlp.get_params()[1],
774                                hidden_size: self.layers[layer].mlp.get_params()[0],
775                                ..Default::default()
776                            },
777                            vb.pp(layer).pp(&mlp).set_dtype(dtype).set_device(device),
778                            &self.mapper.get_comm_for(layer)?,
779                        )?));
780                    }
781                    AnyMoeExpertType::LoraAdapter {
782                        rank,
783                        alpha,
784                        ref target_modules,
785                    } => {
786                        let vb_mlp = vb.pp(layer).pp(&mlp);
787
788                        let gate_proj_delta = if target_modules.contains(&"gate_proj".to_string()) {
789                            Some(get_delta_from_lora_ab!(
790                                vb_mlp,
791                                rank,
792                                alpha,
793                                (hidden_size, intermediate_size),
794                                "gate_proj"
795                            ))
796                        } else {
797                            None
798                        };
799                        let up_proj_delta = if target_modules.contains(&"up_proj".to_string()) {
800                            Some(get_delta_from_lora_ab!(
801                                vb_mlp,
802                                rank,
803                                alpha,
804                                (hidden_size, intermediate_size),
805                                "up_proj"
806                            ))
807                        } else {
808                            None
809                        };
810                        let down_proj_delta = if target_modules.contains(&"down_proj".to_string()) {
811                            Some(get_delta_from_lora_ab!(
812                                vb_mlp,
813                                rank,
814                                alpha,
815                                (intermediate_size, hidden_size),
816                                "down_proj"
817                            ))
818                        } else {
819                            None
820                        };
821
822                        row.push(self.layers[layer].mlp.new_added_delta(vec![
823                            gate_proj_delta,
824                            up_proj_delta,
825                            down_proj_delta,
826                        ])?);
827                    }
828                }
829            }
830        }
831        for (layer, expert) in layers.into_iter().zip(experts) {
832            let mut experts_all = vec![self.layers[layer].mlp.clone()];
833            experts_all.extend(expert);
834            let (dtype, device) = self.layers[layer].mlp.dtype_device();
835            self.layers[layer].mlp = Box::new(MoeMlp::new(
836                experts_all,
837                config.clone(),
838                dtype,
839                &device,
840                layer,
841                gate_vb.as_ref(),
842            )?);
843        }
844        Ok(())
845    }
846    fn amoe_supported(&self) -> bool {
847        true
848    }
849}
mistralrs_core/vision_models/llava/llava_llm/mistral.rs

mistralrs_core/vision_models/llava/llava_llm/
mistral.rs
