# 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 Jais2 model compatible with HuggingFace weights.""" from collections.abc import Iterable import torch from torch import nn from transformers import Jais2Config 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 ReLUSquaredActivation from vllm.model_executor.layers.linear import ( ColumnParallelLinear, 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 ( DEFAULT_VOCAB_PADDING_SIZE, 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 .interfaces import SupportsLoRA, SupportsPP from .utils import ( AutoWeightsLoader, PPMissingLayer, extract_layer_index, is_pp_missing_parameter, make_empty_intermediate_tensors_factory, make_layers, maybe_prefix, ) class Jais2MLP(nn.Module): def __init__( self, hidden_size: int, intermediate_size: int, hidden_act: str, quant_config: QuantizationConfig | None = None, bias: bool = False, prefix: str = "", ) -> None: super().__init__() self.up_proj = ColumnParallelLinear( input_size=hidden_size, output_size=intermediate_size, bias=bias, quant_config=quant_config, prefix=f"{prefix}.up_proj", ) self.down_proj = RowParallelLinear( input_size=intermediate_size, output_size=hidden_size, bias=bias, quant_config=quant_config, prefix=f"{prefix}.down_proj", ) self.act_fn = ReLUSquaredActivation() def forward(self, x): x, _ = self.up_proj(x) x = self.act_fn(x) x, _ = self.down_proj(x) return x class Jais2Attention(nn.Module): def __init__( self, config: Jais2Config, hidden_size: int, num_heads: int, num_kv_heads: int, max_position_embeddings: int = 8192, quant_config: QuantizationConfig | None = None, bias: bool = False, cache_config: CacheConfig | None = None, prefix: str = "", ) -> 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) # MistralConfig has an optional head_dim introduced by Mistral-Nemo self.head_dim = getattr( config, "head_dim", 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, quant_config=quant_config, prefix=f"{prefix}.o_proj", ) is_neox_style = True if quant_config is not None and quant_config.get_name() == "gguf": is_neox_style = False self.rotary_emb = get_rope( self.head_dim, rotary_dim=self.head_dim, max_position=max_position_embeddings, rope_parameters=getattr(config, "rope_parameters", None), is_neox_style=is_neox_style, ) if hasattr(config, "interleaved_sliding_window"): interleaved_sliding_window = config.interleaved_sliding_window if isinstance(interleaved_sliding_window, int): sliding_window = interleaved_sliding_window elif isinstance(interleaved_sliding_window, list): sw_idx = layer_idx % len(interleaved_sliding_window) sliding_window = interleaved_sliding_window[sw_idx] else: raise ValueError( f"{type(interleaved_sliding_window)} is not supported." ) else: sliding_window = None 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, per_layer_sliding_window=sliding_window, 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 class Jais2DecoderLayer(nn.Module): def __init__( self, vllm_config: VllmConfig, config: Jais2Config, prefix: str = "", ) -> 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 ) self.self_attn = Jais2Attention( 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, cache_config=cache_config, prefix=f"{prefix}.self_attn", ) self.mlp = Jais2MLP( 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 = nn.LayerNorm( config.hidden_size, eps=config.layer_norm_eps ) self.post_attention_layernorm = nn.LayerNorm( config.hidden_size, eps=config.layer_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 + 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 + 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 @support_torch_compile class Jais2Model(nn.Module): def __init__( self, vllm_config: VllmConfig, prefix: str = "", layer_type: type[nn.Module] = Jais2DecoderLayer, ): super().__init__() config = vllm_config.model_config.hf_config quant_config = vllm_config.quant_config lora_config = vllm_config.lora_config self.config = config self.quant_config = quant_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 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, org_num_embeddings=config.vocab_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( config=config, vllm_config=vllm_config, prefix=prefix, ), prefix=f"{prefix}.layers", ) if get_pp_group().is_last_rank: self.norm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps) else: self.norm = PPMissingLayer() 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"] for i in range(self.start_layer, self.end_layer): layer = self.layers[i] 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), residual 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"), ] 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 scales for compressed-tensors quantization param = params_dict[scale_name] weight_loader = getattr(param, "weight_loader", default_weight_loader) loaded_weight = loaded_weight[0] weight_loader(param, loaded_weight) loaded_params.add(scale_name) continue if "scale" in name: 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 # Remapping the name of FP8 kv-scale. name = maybe_remap_kv_scale_name(name, params_dict) if name is None: 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 Jais2ForCausalLM(nn.Module, SupportsLoRA, SupportsPP): packed_modules_mapping = { "qkv_proj": ["q_proj", "k_proj", "v_proj"], } embedding_modules = { "embed_tokens": "input_embeddings", "lm_head": "output_embeddings", } def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""): super().__init__() config = vllm_config.model_config.hf_config quant_config = vllm_config.quant_config lora_config = vllm_config.lora_config self.config = config self.lora_config = lora_config self.model = self._init_model( vllm_config=vllm_config, prefix=maybe_prefix(prefix, "model") ) if get_pp_group().is_last_rank: self.unpadded_vocab_size = config.vocab_size if lora_config: self.unpadded_vocab_size += lora_config.lora_extra_vocab_size self.lm_head = ParallelLMHead( self.unpadded_vocab_size, config.hidden_size, org_num_embeddings=config.vocab_size, padding_size=( DEFAULT_VOCAB_PADDING_SIZE # We need bigger padding if using lora for kernel # compatibility if not lora_config else lora_config.lora_vocab_padding_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( self.unpadded_vocab_size, config.vocab_size, logit_scale ) else: self.lm_head = PPMissingLayer() self.make_empty_intermediate_tensors = ( self.model.make_empty_intermediate_tensors ) def _init_model(self, vllm_config: VllmConfig, prefix: str = ""): return Jais2Model(vllm_config=vllm_config, prefix=prefix) 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)