mamba.py

# coding=utf-8
"""PyTorch MAMBA model."""
from typing import Iterable, List, Optional, Tuple

import torch
from torch import nn
from transformers import MambaConfig

from vllm.attention.backends.abstract import AttentionMetadata
from vllm.config import CacheConfig, LoRAConfig, SchedulerConfig
from vllm.distributed import get_tensor_model_parallel_world_size
from vllm.model_executor.layers.layernorm import RMSNorm
from vllm.model_executor.layers.linear import (ColumnParallelLinear,
                                               MergedColumnParallelLinear,
                                               RowParallelLinear)
from vllm.model_executor.layers.logits_processor import LogitsProcessor
from vllm.model_executor.layers.mamba.ops.causal_conv1d import (
    causal_conv1d_fn, causal_conv1d_update)
from vllm.model_executor.layers.mamba.ops.mamba_ssm import (
    selective_scan_fn, selective_state_update)
from vllm.model_executor.layers.quantization.base_config import (
    QuantizationConfig)
from vllm.model_executor.layers.sampler import Sampler, SamplerOutput
from vllm.model_executor.layers.vocab_parallel_embedding import (
    VocabParallelEmbedding)
from vllm.model_executor.model_loader.weight_utils import (
    composed_weight_loader, default_weight_loader, sharded_weight_loader)
from vllm.model_executor.models.interfaces import (HasInnerState,
                                                   IsAttentionFree)
from vllm.model_executor.models.mamba_cache import MambaCacheManager
from vllm.model_executor.sampling_metadata import SamplingMetadata
from vllm.model_executor.utils import set_weight_attrs
from vllm.sequence import IntermediateTensors
from vllm.worker.model_runner import (_BATCH_SIZES_TO_CAPTURE,
                                      _get_graph_batch_size)

KVCache = Tuple[torch.Tensor, torch.Tensor]


# Adapted from transformers.models.mamba.modeling_mamba.MambaMixer
class MambaMixer(nn.Module):
    """
    Compute ∆, A, B, C, and D the state space parameters and compute
    the `contextualized_states`. A, D are input independent
    (see Mamba paper [1] Section 3.5.2 "Interpretation of A"
    for why A isn't selective) ∆, B, C are input-dependent
    (this is a key difference between Mamba and the linear time
    invariant S4, and is why Mamba is called
    **selective** state spaces)
    """

    def __init__(self, config: MambaConfig, layer_idx):
        super().__init__()
        self.config = config
        self.layer_idx = layer_idx
        self.hidden_size = config.hidden_size
        self.ssm_state_size = config.state_size
        self.conv_kernel_size = config.conv_kernel
        self.intermediate_size = config.intermediate_size
        self.time_step_rank = int(config.time_step_rank)

        self.conv1d = ColumnParallelLinear(
            input_size=self.conv_kernel_size,
            output_size=self.intermediate_size,
            bias=config.use_conv_bias,
        )
        # unsqueeze to fit conv1d weights shape into the linear weights shape.
        # Can't do this in `weight_loader` since it already exists in
        # `ColumnParallelLinear` and `set_weight_attrs`
        # doesn't allow to override it
        self.conv1d.weight.data = self.conv1d.weight.data.unsqueeze(1)

        self.in_proj = MergedColumnParallelLinear(self.hidden_size,
                                                  [self.intermediate_size] * 2,
                                                  bias=config.use_bias)
        # selective projection used to make dt, B and C input dependent
        self.x_proj = RowParallelLinear(
            self.intermediate_size,
            self.time_step_rank + self.ssm_state_size * 2,
            bias=False,
        )
        # time step projection (discretization) -
        # In the forward we need to apply dt_proj without the bias,
        # as the bias is added in the selective scan kernel.
        self.dt_proj = ColumnParallelLinear(self.time_step_rank,
                                            self.intermediate_size,
                                            bias=True,
                                            skip_bias_add=True)

        tp_size = get_tensor_model_parallel_world_size()
        self.A = nn.Parameter(
            torch.empty(
                self.intermediate_size // tp_size,
                self.ssm_state_size,
                dtype=torch.float32,
            ))
        self.D = nn.Parameter(torch.ones(self.intermediate_size // tp_size))

        set_weight_attrs(self.D, {"weight_loader": sharded_weight_loader(0)})
        a_weight_loader = composed_weight_loader(
            sharded_weight_loader(0), lambda x: -torch.exp(x.float()))
        set_weight_attrs(self.A, {"weight_loader": a_weight_loader})

        self.out_proj = RowParallelLinear(
            self.intermediate_size,
            self.hidden_size,
            bias=config.use_bias,
            input_is_parallel=True,
        )
        self.activation = config.hidden_act

    def forward(self, hidden_states: torch.Tensor,
                attn_metadata: AttentionMetadata, conv_state: torch.Tensor,
                ssm_state: torch.Tensor):

        # 1. Gated MLP's linear projection
        projected_states = self.in_proj(hidden_states)[0].transpose(-2, -1)
        hidden_states, gate = projected_states.chunk(2, dim=-2)

        # 2. Convolution sequence transformation
        conv_weights = self.conv1d.weight.view(self.conv1d.weight.size(0),
                                               self.conv1d.weight.size(2))

        if attn_metadata.query_start_loc is not None \
            and attn_metadata.context_lens_tensor is not None:
            # |---------- N-1 iteration --------|
            # |---------------- N iteration ---------------------|
            # |- tokenA -|......................|-- newTokens ---|
            # |---------- context_len ----------|
            # |-------------------- seq_len ---------------------|
            #                                   |-- query_len ---|
            hidden_states = causal_conv1d_fn(
                hidden_states,
                conv_weights,
                self.conv1d.bias,
                activation=self.activation,
                conv_states=conv_state,
                has_initial_state=attn_metadata.context_lens_tensor > 0,
                query_start_loc=attn_metadata.query_start_loc)
        else:
            hidden_states = causal_conv1d_update(
                hidden_states.transpose(0, 1),
                conv_state,
                conv_weights,
                self.conv1d.bias,
                self.activation,
            )
            hidden_states = hidden_states.transpose(0, 1)

        # 3. State Space Model sequence transformation
        # 3.a. input varying initialization of time_step, B and C
        ssm_parameters = self.x_proj(hidden_states.transpose(-2, -1))[0]

        time_step, B, C = torch.split(
            ssm_parameters,
            [self.time_step_rank, self.ssm_state_size, self.ssm_state_size],
            dim=-1,
        )

        # Note that Jamba normalizes B, C, and time_step here but Mamba doesn't.

        discrete_time_step = self.dt_proj(time_step)[0].transpose(-2, -1)
        # 3.c perform the recurrence y ← SSM(A, B, C)(x)
        time_proj_bias = (self.dt_proj.bias.float() if hasattr(
            self.dt_proj, "bias") else None)

        if attn_metadata.query_start_loc is not None \
            and attn_metadata.context_lens_tensor is not None:
            scan_outputs = selective_scan_fn(
                hidden_states,
                ssm_state,
                discrete_time_step,
                self.A,
                B.transpose(-2, -1),
                C.transpose(-2, -1),
                self.D.float(),
                gate,
                time_proj_bias,
                delta_softplus=True,
                has_initial_state=attn_metadata.context_lens_tensor > 0,
                query_start_loc=attn_metadata.query_start_loc)
        else:
            scan_outputs = selective_state_update(
                ssm_state,
                hidden_states.transpose(0, 1),
                discrete_time_step.transpose(0, 1),
                self.A,
                B,
                C,
                self.D,
                gate.transpose(0, 1),
                time_proj_bias,
                dt_softplus=True,
            )
            scan_outputs = scan_outputs.transpose(0, 1)

        # 4. Final linear projection
        contextualized_states = self.out_proj(scan_outputs.transpose(-2,
                                                                     -1))[0]
        return contextualized_states


class MambaDecoderLayer(nn.Module):

    def __init__(self,
                 config: MambaConfig,
                 layer_idx: int,
                 cache_config: Optional[CacheConfig] = None,
                 quant_config: Optional[QuantizationConfig] = None) -> None:
        super().__init__()
        self.layer_idx = layer_idx
        self.config = config
        self.mixer = MambaMixer(config, layer_idx)

        self.norm = RMSNorm(config.hidden_size, eps=config.layer_norm_epsilon)
        self.pre_ff_layernorm = RMSNorm(config.hidden_size,
                                        eps=config.layer_norm_epsilon)

    def forward(
        self,
        hidden_states: torch.Tensor,
        attn_metadata: AttentionMetadata,
        residual: Optional[torch.Tensor],
        conv_state: torch.Tensor,
        ssm_state: torch.Tensor,
        **kwargs,
    ):
        if residual is None:
            residual = hidden_states
            hidden_states = self.norm(hidden_states)
        else:
            hidden_states, residual = self.norm(hidden_states, residual)

        hidden_states = self.mixer(hidden_states, attn_metadata, conv_state,
                                   ssm_state)
        return hidden_states, residual


class MambaModel(nn.Module):

    def __init__(
        self,
        config: MambaConfig,
        quant_config: Optional[QuantizationConfig] = None,
        cache_config: Optional[CacheConfig] = None,
        lora_config: Optional[LoRAConfig] = None,
    ) -> None:
        super().__init__()
        self.config = config
        self.padding_idx = config.pad_token_id
        lora_vocab = ((lora_config.lora_extra_vocab_size *
                       (lora_config.max_loras or 1)) if lora_config else 0)
        self.vocab_size = config.vocab_size + lora_vocab
        self.org_vocab_size = config.vocab_size

        self.embeddings = VocabParallelEmbedding(
            self.vocab_size,
            config.hidden_size,
            org_num_embeddings=config.vocab_size,
        )

        decoder_layers = []
        for i in range(config.num_hidden_layers):
            decoder_layers.append(
                MambaDecoderLayer(config,
                                  layer_idx=i,
                                  cache_config=cache_config,
                                  quant_config=quant_config))
        self.layers = nn.ModuleList(decoder_layers)
        self.norm_f = RMSNorm(config.hidden_size,
                              eps=config.layer_norm_epsilon)

    def forward(
        self,
        input_ids: torch.Tensor,
        positions: torch.Tensor,
        attn_metadata: AttentionMetadata,
        conv_state: torch.Tensor,
        ssm_state: torch.Tensor,
    ) -> torch.Tensor:
        hidden_states = self.embeddings(input_ids)
        residual = None

        for i in range(len(self.layers)):
            layer = self.layers[i]
            current_ssm_state = ssm_state[i]
            current_conv_state = conv_state[i]

            hidden_states, residual = layer(
                positions=positions,
                hidden_states=hidden_states,
                attn_metadata=attn_metadata,
                residual=residual,
                conv_state=current_conv_state,
                ssm_state=current_ssm_state,
            )
        hidden_states, _ = self.norm_f(hidden_states, residual)

        return hidden_states


class MambaForCausalLM(nn.Module, HasInnerState, IsAttentionFree):

    def __init__(
        self,
        config: MambaConfig,
        cache_config: Optional[CacheConfig] = None,
        quant_config: Optional[QuantizationConfig] = None,
        lora_config: Optional[LoRAConfig] = None,
        scheduler_config: Optional[SchedulerConfig] = None,
    ) -> None:
        assert not cache_config.enable_prefix_caching, \
            "Mamba does not support prefix caching"

        super().__init__()
        self.config = config
        self.scheduler_config = scheduler_config
        self.backbone = MambaModel(config,
                                   cache_config=cache_config,
                                   quant_config=quant_config,
                                   lora_config=lora_config)
        self.unpadded_vocab_size = config.vocab_size
        if lora_config:
            self.unpadded_vocab_size += lora_config.lora_extra_vocab_size

        self.lm_head = self.backbone.embeddings

        # Used to track and store by the Mamba cache between steps.
        self.mamba_cache: Optional[MambaCacheManager] = None

        self.logits_processor = LogitsProcessor(self.unpadded_vocab_size,
                                                config.vocab_size)
        self.sampler = Sampler()

    def forward(self,
                input_ids: torch.Tensor,
                positions: torch.Tensor,
                kv_caches: List[KVCache],
                attn_metadata: AttentionMetadata,
                intermediate_tensors: Optional[IntermediateTensors] = None,
                **kwargs):
        if self.mamba_cache is None:
            max_batch_size = (_get_graph_batch_size(
                self.scheduler_config.max_num_seqs) if self.scheduler_config
                              else max(_BATCH_SIZES_TO_CAPTURE) + 2)
            self.mamba_cache = MambaCacheManager(
                self.lm_head.weight.dtype, self.config.num_hidden_layers,
                max_batch_size, *self._get_mamba_cache_shape())

        mamba_cache_tensors = self.mamba_cache.current_run_tensors(
            input_ids, attn_metadata, **kwargs)

        hidden_states = self.backbone(input_ids, positions, attn_metadata,
                                      mamba_cache_tensors[0],
                                      mamba_cache_tensors[1])

        return hidden_states

    def copy_inputs_before_cuda_graphs(self, input_buffers, **kwargs):
        return self.mamba_cache.copy_inputs_before_cuda_graphs(
            input_buffers, **kwargs)

    def get_seqlen_agnostic_capture_inputs(self, batch_size: int):
        return self.mamba_cache.get_seqlen_agnostic_capture_inputs(batch_size)

    def _get_mamba_cache_shape(
            self) -> Tuple[Tuple[int, int], Tuple[int, int]]:
        world_size = get_tensor_model_parallel_world_size()
        conv_state_shape = (
            self.config.intermediate_size // world_size,
            self.config.conv_kernel - 1,
        )
        temporal_state_shape = (
            self.config.intermediate_size // world_size,
            self.config.state_size,
        )
        return conv_state_shape, temporal_state_shape

    def compute_logits(self, hidden_states: torch.Tensor,
                       sampling_metadata: SamplingMetadata) -> torch.Tensor:
        logits = self.logits_processor(self.lm_head, hidden_states,
                                       sampling_metadata)
        return logits

    def sample(
        self,
        logits: Optional[torch.Tensor],
        sampling_metadata: SamplingMetadata,
    ) -> Optional[SamplerOutput]:
        next_tokens = self.sampler(logits, sampling_metadata)
        return next_tokens

    def load_weights(self, weights: Iterable[Tuple[str, torch.Tensor]]):
        params_dict = dict(self.named_parameters())
        for name, loaded_weight in weights:
            if "A_log" in name:
                name = name.replace("A_log", "A")

            # Skip loading extra bias for GPTQ models.
            if name.endswith(".bias") and name not in params_dict:
                continue

            param = params_dict[name]
            weight_loader = getattr(param, "weight_loader",
                                    default_weight_loader)
            weight_loader(param, loaded_weight)