glm4.py

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

# Copyright 2025 The Zhipu AI team.
# 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 GLM-4-0414 model compatible with HuggingFace weights."""

import os
from collections.abc import Iterable

import torch
from torch import nn
from transformers import Glm4Config

from vllm.attention.backends.abstract import AttentionType
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.layernorm import RMSNorm
from vllm.model_executor.layers.linear import 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
from vllm.sequence import IntermediateTensors

from .interfaces import SupportsLoRA, SupportsPP
from .llama import LlamaMLP as Glm4MLP
from .llama import LlamaModel
from .utils import AutoWeightsLoader, PPMissingLayer, maybe_prefix

from vllm.utils import W8a8GetCacheJSON
from vllm import _custom_ops as ops
from vllm.model_executor.utils import pad_weight, gemm_bank_conf

class Glm4Attention(nn.Module):
    def __init__(
        self,
        config: Glm4Config,
        hidden_size: int,
        num_heads: int,
        num_kv_heads: int,
        max_position: int = 4096 * 32,
        head_dim: int | None = None,
        qkv_bias: bool = False,
        cache_config: CacheConfig | None = None,
        quant_config: QuantizationConfig | None = None,
        prefix: str = "",
        attn_type: str = AttentionType.DECODER,
    ) -> None:
        super().__init__()
        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
        config.rope_parameters.setdefault("partial_rotary_factor", 0.5)
        self.num_kv_heads = max(1, self.total_num_kv_heads // tp_size)
        self.head_dim = head_dim or 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.qkv_proj = QKVParallelLinear(
            hidden_size,
            self.head_dim,
            self.total_num_heads,
            self.total_num_kv_heads,
            bias=qkv_bias,
            quant_config=quant_config,
            prefix=f"{prefix}.qkv_proj",
        )
        self.o_proj = RowParallelLinear(
            self.total_num_heads * self.head_dim,
            hidden_size,
            bias=False,
            quant_config=quant_config,
            prefix=f"{prefix}.o_proj",
        )
        self.rotary_emb = get_rope(
            self.head_dim,
            max_position=max_position,
            rope_parameters=config.rope_parameters,
            is_neox_style=False,
        )
        self.attn = Attention(
            self.num_heads,
            self.head_dim,
            self.scaling,
            num_kv_heads=self.num_kv_heads,
            cache_config=cache_config,
            quant_config=quant_config,
            prefix=f"{prefix}.attn",
            attn_type=attn_type,
        )

    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


class Glm4DecoderLayer(nn.Module):
    def __init__(
        self,
        vllm_config: VllmConfig,
        prefix: str = "",
        config: Glm4Config | None = None,
    ) -> None:
        super().__init__()

        config = config or vllm_config.model_config.hf_config
        cache_config = vllm_config.cache_config
        quant_config = vllm_config.quant_config

        self.hidden_size = config.hidden_size

        self.self_attn = Glm4Attention(
            config=config,
            hidden_size=self.hidden_size,
            num_heads=config.num_attention_heads,
            max_position=config.max_position_embeddings,
            num_kv_heads=config.num_key_value_heads,
            qkv_bias=getattr(config, "attention_bias", False),
            head_dim=getattr(config, "head_dim", None),
            cache_config=cache_config,
            quant_config=quant_config,
            prefix=f"{prefix}.self_attn",
            attn_type=AttentionType.DECODER,
        )
        self.mlp = Glm4MLP(
            hidden_size=self.hidden_size,
            intermediate_size=config.intermediate_size,
            hidden_act=config.hidden_act,
            quant_config=quant_config,
            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
        )
        self.post_self_attn_layernorm = RMSNorm(
            config.hidden_size, eps=config.rms_norm_eps
        )
        self.post_mlp_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,
        )

        hidden_states = self.post_self_attn_layernorm(hidden_states)

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

        return hidden_states, residual


ALL_DECODER_LAYER_TYPES = {
    "attention": Glm4DecoderLayer,
}


@support_torch_compile(
    dynamic_arg_dims={
        "input_ids": 0,
        "positions": -1,
        "intermediate_tensors": 0,
        "inputs_embeds": 0,
    }
)
class Glm4Model(LlamaModel):
    def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
        super().__init__(
            vllm_config=vllm_config, prefix=prefix, layer_type=Glm4DecoderLayer
        )


class Glm4ForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
    packed_modules_mapping = {
        "qkv_proj": [
            "q_proj",
            "k_proj",
            "v_proj",
        ],
        "gate_up_proj": [
            "gate_proj",
            "up_proj",
        ],
    }

    def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
        super().__init__()
        config = vllm_config.model_config.hf_config
        quant_config = vllm_config.quant_config

        self.config = config

        self.quant_config = quant_config
        self.model = Glm4Model(
            vllm_config=vllm_config, prefix=maybe_prefix(prefix, "model")
        )

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

        self.logits_processor = LogitsProcessor(config.vocab_size)

        self.make_empty_intermediate_tensors = (
            self.model.make_empty_intermediate_tensors
        )
        self.quant_method = None
        if quant_config is not None:
            self.quant_method=quant_config.get_name()
            self.quant_config=quant_config
            
        self.tritonsingleton= W8a8GetCacheJSON()      
        self.use_llama_nn = os.environ.get('LLAMA_NN') == '1'
        # self.use_lm_nn = os.environ.get('LM_NN') == '1'
        self.use_gemm_pad = os.environ.get('GEMM_PAD') == '1'
        self.use_fa_pad = os.environ.get('FA_PAD') == '1'
        self.use_awq_pad = os.environ.get('AWQ_PAD') == '1'
        self.w8a8_strategy=int(os.getenv('W8A8_SUPPORT_METHODS', '1'))

    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:
        hidden_states = self.model(
            input_ids, positions, intermediate_tensors, inputs_embeds
        )
        return hidden_states

    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)