llama.py

# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project

# Adapted from
# https://github.com/huggingface/transformers/blob/v4.28.0/src/transformers/models/llama/modeling_llama.py
# Copyright 2023 The vLLM team.
# Copyright 2022 EleutherAI and the HuggingFace Inc. team. All rights reserved.
#
# This code is based on EleutherAI's GPT-NeoX library and the GPT-NeoX
# and OPT implementations in this library. It has been modified from its
# original forms to accommodate minor architectural differences compared
# to GPT-NeoX and OPT used by the Meta AI team that trained the model.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Inference-only LLaMA model compatible with HuggingFace weights."""

from collections.abc import Iterable
from itertools import islice

import torch
from torch import nn
from transformers import LlamaConfig

from vllm.attention.layer import Attention
from vllm.compilation.decorators import support_torch_compile
from vllm.config import CacheConfig, VllmConfig
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
from vllm.model_executor.layers.activation import SiluAndMul
from vllm.model_executor.layers.attention.encoder_only_attention import (
    EncoderOnlyAttention,
)
from vllm.model_executor.layers.layernorm import RMSNorm
from vllm.model_executor.layers.linear import (
    MergedColumnParallelLinear,
    QKVParallelLinear,
    RowParallelLinear,
)
from vllm.model_executor.layers.logits_processor import LogitsProcessor
from vllm.model_executor.layers.quantization import QuantizationConfig
from vllm.model_executor.layers.rotary_embedding import get_rope
from vllm.model_executor.layers.vocab_parallel_embedding import (
    ParallelLMHead,
    VocabParallelEmbedding,
)
from vllm.model_executor.model_loader.weight_utils import (
    default_weight_loader,
    maybe_remap_kv_scale_name,
)
from vllm.sequence import IntermediateTensors
from vllm.v1.attention.backend import AttentionType

from .adapters import as_embedding_model, as_seq_cls_model
from .interfaces import (
    SupportsEagle,
    SupportsEagle3,
    SupportsLoRA,
    SupportsPP,
)
from .utils import (
    AutoWeightsLoader,
    PPMissingLayer,
    extract_layer_index,
    is_pp_missing_parameter,
    make_empty_intermediate_tensors_factory,
    make_layers,
    maybe_prefix,
)


class LlamaMLP(nn.Module):
    def __init__(
        self,
        hidden_size: int,
        intermediate_size: int,
        hidden_act: str,
        quant_config: QuantizationConfig | None = None,
        bias: bool = False,
        prefix: str = "",
        reduce_results: bool = True,
        disable_tp: bool = False,
    ) -> None:
        super().__init__()
        self.gate_up_proj = MergedColumnParallelLinear(
            input_size=hidden_size,
            output_sizes=[intermediate_size] * 2,
            bias=bias,
            quant_config=quant_config,
            disable_tp=disable_tp,
            prefix=f"{prefix}.gate_up_proj",
        )
        self.down_proj = RowParallelLinear(
            input_size=intermediate_size,
            output_size=hidden_size,
            bias=bias,
            quant_config=quant_config,
            reduce_results=reduce_results,
            disable_tp=disable_tp,
            prefix=f"{prefix}.down_proj",
        )
        if hidden_act != "silu":
            raise ValueError(
                f"Unsupported activation: {hidden_act}. Only silu is supported for now."
            )
        self.act_fn = SiluAndMul()

    def forward(self, x):
        x, _ = self.gate_up_proj(x)
        x = self.act_fn(x)
        x, _ = self.down_proj(x)
        return x


class LlamaAttention(nn.Module):
    def __init__(
        self,
        config: LlamaConfig,
        hidden_size: int,
        num_heads: int,
        num_kv_heads: int,
        max_position_embeddings: int = 8192,
        quant_config: QuantizationConfig | None = None,
        bias: bool = False,
        bias_o_proj: bool = False,
        cache_config: CacheConfig | None = None,
        prefix: str = "",
        attn_type: str = AttentionType.DECODER,
    ) -> None:
        super().__init__()
        layer_idx = extract_layer_index(prefix)
        self.hidden_size = hidden_size
        tp_size = get_tensor_model_parallel_world_size()
        self.total_num_heads = num_heads
        assert self.total_num_heads % tp_size == 0
        self.num_heads = self.total_num_heads // tp_size
        self.total_num_kv_heads = num_kv_heads
        if self.total_num_kv_heads >= tp_size:
            # Number of KV heads is greater than TP size, so we partition
            # the KV heads across multiple tensor parallel GPUs.
            assert self.total_num_kv_heads % tp_size == 0
        else:
            # Number of KV heads is less than TP size, so we replicate
            # the KV heads across multiple tensor parallel GPUs.
            assert tp_size % self.total_num_kv_heads == 0
        self.num_kv_heads = max(1, self.total_num_kv_heads // tp_size)

        head_dim = getattr(config, "head_dim", None)
        self.head_dim = head_dim or self.hidden_size // self.total_num_heads
        self.q_size = self.num_heads * self.head_dim
        self.kv_size = self.num_kv_heads * self.head_dim
        self.scaling = self.head_dim**-0.5
        self.max_position_embeddings = max_position_embeddings

        self.qkv_proj = QKVParallelLinear(
            hidden_size=hidden_size,
            head_size=self.head_dim,
            total_num_heads=self.total_num_heads,
            total_num_kv_heads=self.total_num_kv_heads,
            bias=bias,
            quant_config=quant_config,
            prefix=f"{prefix}.qkv_proj",
        )

        self.o_proj = RowParallelLinear(
            input_size=self.total_num_heads * self.head_dim,
            output_size=hidden_size,
            bias=bias_o_proj,
            quant_config=quant_config,
            prefix=f"{prefix}.o_proj",
        )

        self._init_rotary_emb(config, quant_config=quant_config)

        sliding_window = None
        if layer_types := getattr(config, "layer_types", None):
            # Fix for Eagle3 compatibility:
            # for draft models, subtract target layer count
            # to get draft-relative layer index starting from 0
            if hasattr(config, "target_layer_count"):
                # This is a draft model,
                # adjust layer_idx to be relative to draft layers
                effective_layer_idx = layer_idx - config.target_layer_count
            else:
                # This is a target model, use layer_idx directly
                effective_layer_idx = layer_idx
            assert effective_layer_idx < len(layer_types), (
                f"effective_layer_idx: {effective_layer_idx} "
                f"is out of bounds for layer_types: {layer_types}"
            )

            is_sliding = layer_types[effective_layer_idx] == "sliding_attention"
            if is_sliding:
                sliding_window = config.sliding_window

        attn_cls = (
            EncoderOnlyAttention
            if attn_type == AttentionType.ENCODER_ONLY
            else Attention
        )

        self.attn = attn_cls(
            self.num_heads,
            self.head_dim,
            self.scaling,
            num_kv_heads=self.num_kv_heads,
            cache_config=cache_config,
            quant_config=quant_config,
            per_layer_sliding_window=sliding_window,
            attn_type=attn_type,
            prefix=f"{prefix}.attn",
        )

    def forward(
        self,
        positions: torch.Tensor,
        hidden_states: torch.Tensor,
    ) -> torch.Tensor:
        qkv, _ = self.qkv_proj(hidden_states)
        q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
        q, k = self.rotary_emb(positions, q, k)
        attn_output = self.attn(q, k, v)
        output, _ = self.o_proj(attn_output)
        return output

    def _init_rotary_emb(
        self,
        config: LlamaConfig,
        quant_config: QuantizationConfig | None,
    ) -> None:
        is_neox_style = True
        is_gguf = quant_config and quant_config.get_name() == "gguf"
        if is_gguf and config.model_type == "llama":
            is_neox_style = False

        self.rotary_emb = get_rope(
            self.head_dim,
            max_position=self.max_position_embeddings,
            rope_parameters=getattr(config, "rope_parameters", None),
            is_neox_style=is_neox_style,
        )


class LlamaDecoderLayer(nn.Module):
    def __init__(
        self,
        vllm_config: VllmConfig,
        prefix: str = "",
        config: LlamaConfig | None = None,
        attn_layer_type: type[nn.Module] = LlamaAttention,
    ) -> None:
        super().__init__()

        config = config or vllm_config.model_config.hf_config
        cache_config = vllm_config.cache_config
        quant_config = self.get_quant_config(vllm_config)

        self.hidden_size = config.hidden_size
        max_position_embeddings = getattr(config, "max_position_embeddings", 8192)
        # Support abacusai/Smaug-72B-v0.1 with attention_bias
        # Support internlm/internlm-7b with bias
        attention_bias = getattr(config, "attention_bias", False) or getattr(
            config, "bias", False
        )
        bias_o_proj = attention_bias
        # support internlm/internlm3-8b with qkv_bias
        if hasattr(config, "qkv_bias"):
            attention_bias = config.qkv_bias

        # By default, Llama uses causal attention as it is a decoder-only model.
        # You can override the HF config with `is_causal=False` to enable
        # bidirectional attention, which is used in some embedding models
        # (e.g. parasail-ai/GritLM-7B-vllm)
        if getattr(config, "is_causal", True):
            attn_type = AttentionType.DECODER
        else:
            attn_type = AttentionType.ENCODER_ONLY

        self.self_attn = attn_layer_type(
            config=config,
            hidden_size=self.hidden_size,
            num_heads=config.num_attention_heads,
            num_kv_heads=getattr(
                config, "num_key_value_heads", config.num_attention_heads
            ),
            max_position_embeddings=max_position_embeddings,
            quant_config=quant_config,
            bias=attention_bias,
            bias_o_proj=bias_o_proj,
            cache_config=cache_config,
            prefix=f"{prefix}.self_attn",
            attn_type=attn_type,
        )
        self.mlp = LlamaMLP(
            hidden_size=self.hidden_size,
            intermediate_size=config.intermediate_size,
            hidden_act=config.hidden_act,
            quant_config=quant_config,
            bias=getattr(config, "mlp_bias", False),
            prefix=f"{prefix}.mlp",
        )
        self.input_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
        self.post_attention_layernorm = RMSNorm(
            config.hidden_size, eps=config.rms_norm_eps
        )

    def forward(
        self,
        positions: torch.Tensor,
        hidden_states: torch.Tensor,
        residual: torch.Tensor | None,
    ) -> tuple[torch.Tensor, torch.Tensor]:
        # Self Attention
        if residual is None:
            residual = hidden_states
            hidden_states = self.input_layernorm(hidden_states)
        else:
            hidden_states, residual = self.input_layernorm(hidden_states, residual)
        hidden_states = self.self_attn(positions=positions, hidden_states=hidden_states)

        # Fully Connected
        hidden_states, residual = self.post_attention_layernorm(hidden_states, residual)
        hidden_states = self.mlp(hidden_states)
        return hidden_states, residual

    def get_quant_config(self, vllm_config: VllmConfig) -> QuantizationConfig | None:
        """Get quantization config for this layer. Override in subclasses."""
        return vllm_config.quant_config


def llama_model_invariants(
    input_ids, positions, intermediate_tensors=None, inputs_embeds=None
):
    """Shape invariants for Llama model compilation, those are translated to
    runtime assertions for unbacked dynamic shapes and are compiled away for
    backed"""
    if input_ids is not None:
        torch._check(positions.size()[0] == input_ids.size()[0])


@support_torch_compile(
    # TODO[#32068]: Investigate recompilation
    # mark_unbacked_dims={"input_ids": 0},
    shape_invariants=llama_model_invariants
)
class LlamaModel(nn.Module):
    def __init__(
        self,
        *,
        vllm_config: VllmConfig,
        prefix: str = "",
        layer_type: type[nn.Module] = LlamaDecoderLayer,
    ):
        super().__init__()

        config = vllm_config.model_config.hf_config
        quant_config = vllm_config.quant_config

        self.config = config
        self.quant_config = quant_config

        self.vocab_size = config.vocab_size

        if get_pp_group().is_first_rank or (
            config.tie_word_embeddings and get_pp_group().is_last_rank
        ):
            self.embed_tokens = VocabParallelEmbedding(
                self.vocab_size,
                config.hidden_size,
                quant_config=quant_config,
            )
        else:
            self.embed_tokens = PPMissingLayer()
        self.start_layer, self.end_layer, self.layers = make_layers(
            config.num_hidden_layers,
            lambda prefix: layer_type(vllm_config=vllm_config, prefix=prefix),
            prefix=f"{prefix}.layers",
        )
        if get_pp_group().is_last_rank:
            self.norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
        else:
            self.norm = PPMissingLayer()

        self.aux_hidden_state_layers = tuple[int, ...]()

        self.make_empty_intermediate_tensors = make_empty_intermediate_tensors_factory(
            ["hidden_states", "residual"], config.hidden_size
        )

    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
        return self.embed_tokens(input_ids)

    def forward(
        self,
        input_ids: torch.Tensor | None,
        positions: torch.Tensor,
        intermediate_tensors: IntermediateTensors | None,
        inputs_embeds: torch.Tensor | None = None,
    ) -> torch.Tensor | IntermediateTensors | tuple[torch.Tensor, list[torch.Tensor]]:
        if get_pp_group().is_first_rank:
            if inputs_embeds is not None:
                hidden_states = inputs_embeds
            else:
                hidden_states = self.embed_input_ids(input_ids)
            residual = None
        else:
            assert intermediate_tensors is not None
            hidden_states = intermediate_tensors["hidden_states"]
            residual = intermediate_tensors["residual"]

        aux_hidden_states = []
        for idx, layer in enumerate(
            islice(self.layers, self.start_layer, self.end_layer)
        ):
            if idx in self.aux_hidden_state_layers:
                aux_hidden_states.append(hidden_states + residual)
            hidden_states, residual = layer(positions, hidden_states, residual)

        if not get_pp_group().is_last_rank:
            return IntermediateTensors(
                {"hidden_states": hidden_states, "residual": residual}
            )

        hidden_states, _ = self.norm(hidden_states, residual)

        if len(aux_hidden_states) > 0:
            return hidden_states, aux_hidden_states
        return hidden_states

    def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]:
        stacked_params_mapping = [
            # (param_name, shard_name, shard_id)
            (".qkv_proj", ".q_proj", "q"),
            (".qkv_proj", ".k_proj", "k"),
            (".qkv_proj", ".v_proj", "v"),
            (".gate_up_proj", ".gate_proj", 0),
            (".gate_up_proj", ".up_proj", 1),
        ]
        params_dict = dict(self.named_parameters())
        loaded_params: set[str] = set()
        for name, loaded_weight in weights:
            if "rotary_emb.inv_freq" in name:
                continue
            if "rotary_emb.cos_cached" in name or "rotary_emb.sin_cached" in name:
                # Models trained using ColossalAI may include these tensors in
                # the checkpoint. Skip them.
                continue
            if self.quant_config is not None and (
                scale_name := self.quant_config.get_cache_scale(name)
            ):
                # Loading kv cache quantization scales
                param = params_dict[scale_name]
                weight_loader = getattr(param, "weight_loader", default_weight_loader)
                loaded_weight = (
                    loaded_weight if loaded_weight.dim() == 0 else loaded_weight[0]
                )
                weight_loader(param, loaded_weight)
                loaded_params.add(scale_name)
                continue
            if "scale" in name or "zero_point" in name:
                # Remapping the name of FP8 kv-scale or zero point.
                name = maybe_remap_kv_scale_name(name, params_dict)
                if name is None:
                    continue
            for param_name, weight_name, shard_id in stacked_params_mapping:
                if weight_name not in name:
                    continue
                name = name.replace(weight_name, param_name)
                # Skip loading extra bias for GPTQ models.
                if name.endswith(".bias") and name not in params_dict:
                    continue

                if is_pp_missing_parameter(name, self):
                    continue

                param = params_dict[name]
                weight_loader = param.weight_loader
                weight_loader(param, loaded_weight, shard_id)
                break
            else:
                # Skip loading extra bias for GPTQ models.
                if name.endswith(".bias") and name not in params_dict:
                    continue

                if is_pp_missing_parameter(name, self):
                    continue

                param = params_dict[name]
                weight_loader = getattr(param, "weight_loader", default_weight_loader)
                weight_loader(param, loaded_weight)
            loaded_params.add(name)
        return loaded_params


class LlamaForCausalLM(
    nn.Module, SupportsLoRA, SupportsPP, SupportsEagle, SupportsEagle3
):
    packed_modules_mapping = {
        "qkv_proj": ["q_proj", "k_proj", "v_proj"],
        "gate_up_proj": ["gate_proj", "up_proj"],
    }

    # LoRA specific attributes
    embedding_modules = {
        "embed_tokens": "input_embeddings",
        "lm_head": "output_embeddings",
    }

    def __init__(
        self,
        *,
        vllm_config: VllmConfig,
        prefix: str = "",
        layer_type: type[nn.Module] = LlamaDecoderLayer,
    ):
        super().__init__()
        config = vllm_config.model_config.hf_config
        quant_config = vllm_config.quant_config
        self.config = config

        self.model = self._init_model(
            vllm_config=vllm_config,
            prefix=maybe_prefix(prefix, "model"),
            layer_type=layer_type,
        )

        if get_pp_group().is_last_rank:
            self.lm_head = ParallelLMHead(
                config.vocab_size,
                config.hidden_size,
                quant_config=quant_config,
                prefix=maybe_prefix(prefix, "lm_head"),
            )
            if config.tie_word_embeddings:
                self.lm_head = self.lm_head.tie_weights(self.model.embed_tokens)

            logit_scale = getattr(config, "logit_scale", 1.0)
            self.logits_processor = LogitsProcessor(
                config.vocab_size, scale=logit_scale
            )
        else:
            self.lm_head = PPMissingLayer()

        self.make_empty_intermediate_tensors = (
            self.model.make_empty_intermediate_tensors
        )

    def set_aux_hidden_state_layers(self, layers: tuple[int, ...]) -> None:
        self.model.aux_hidden_state_layers = layers

    def get_eagle3_aux_hidden_state_layers(self) -> tuple[int, ...]:
        """Override to return default layers for Llama

        Note: The GPU model runner will override this with layers from
        the speculative config if available, providing dynamic configuration.
        """
        num_layers = len(self.model.layers)
        return (2, num_layers // 2, num_layers - 3)

    def _init_model(
        self,
        vllm_config: VllmConfig,
        prefix: str = "",
        layer_type: type[nn.Module] = LlamaDecoderLayer,
    ):
        return LlamaModel(vllm_config=vllm_config, prefix=prefix, layer_type=layer_type)

    def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
        return self.model.embed_input_ids(input_ids)

    def forward(
        self,
        input_ids: torch.Tensor,
        positions: torch.Tensor,
        intermediate_tensors: IntermediateTensors | None = None,
        inputs_embeds: torch.Tensor | None = None,
    ) -> torch.Tensor | IntermediateTensors:
        model_output = self.model(
            input_ids, positions, intermediate_tensors, inputs_embeds
        )
        return model_output

    def compute_logits(
        self,
        hidden_states: torch.Tensor,
    ) -> torch.Tensor | None:
        logits = self.logits_processor(self.lm_head, hidden_states)
        return logits

    def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]:
        loader = AutoWeightsLoader(
            self,
            skip_prefixes=(["lm_head."] if self.config.tie_word_embeddings else None),
        )
        return loader.load_weights(weights)


class LlamaBidirectionalForSequenceClassification(as_seq_cls_model(LlamaForCausalLM)):
    # This class sets the correct attention type and pooling type
    # through LlamaBidirectionalConfig.
    pass


class LlamaBidirectionalModel(as_embedding_model(LlamaForCausalLM)):
    # This class sets the correct attention type and pooling type
    # through LlamaBidirectionalConfig.
    pass