from typing import Callable, List, Optional, Tuple, Union import math import torch import torch.nn as nn from transformers.models.llama import LlamaForCausalLM, LlamaConfig from transformers.models.diffllama import DiffLlamaForCausalLM, DiffLlamaConfig from transformers.utils import logging from transformers.modeling_utils import ALL_ATTENTION_FUNCTIONS logger = logging.get_logger(__name__) class DiffLlamaForCausalLMv0(DiffLlamaForCausalLM): def __init__(self, config: LlamaConfig): super().__init__(config) for layer_i, layer in enumerate(self.model.layers): hidden_dim = config.hidden_size n_heads = config.num_attention_heads depth = layer_i # or pass 0 layer.self_attn = DiffLlamaSdpaAttention( config, layer_idx=layer_i ) print("# Initializing Fixed Lambda DiffLlama") def lambda_init_fn(layer_idx): return 0.8 - 0.6 * math.exp(-0.3 * layer_idx) class DiffLlamaAttention(nn.Module): """Multi-headed attention from 'Attention Is All You Need' paper""" def __init__(self, config: DiffLlamaConfig, layer_idx: Optional[int] = None): super().__init__() self.config = config self.layer_idx = layer_idx if layer_idx is None: logger.warning_once( f"Instantiating {self.__class__.__name__} without passing a `layer_idx` is not recommended and will " "lead to errors during the forward call if caching is used. Please make sure to provide a `layer_idx` " "when creating this class." ) self.attention_dropout = config.attention_dropout self.hidden_size = config.hidden_size self.num_heads = config.num_attention_heads self.head_dim = getattr(config, "head_dim", self.hidden_size // self.num_heads) self.num_key_value_heads = config.num_key_value_heads self.num_key_value_groups = self.num_heads // self.num_key_value_heads # under this are not used self.max_position_embeddings = config.max_position_embeddings self.rope_theta = config.rope_theta self.is_causal = True self.q_proj = nn.Linear(self.hidden_size, self.num_heads * self.head_dim, bias=config.attention_bias) self.k_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=config.attention_bias) self.v_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=config.attention_bias) self.o_proj = nn.Linear(self.num_heads * self.head_dim, self.hidden_size, bias=config.attention_bias) self.lambda_init = lambda_init_fn(layer_idx) self.groupnorm = nn.RMSNorm(2 * self.head_dim, eps=config.rms_norm_eps, elementwise_affine=False) # self.collected_attn_logits = None def _store_attn_logits(self, attn_logits): top_vals, _ = torch.topk(attn_logits.reshape(-1), k=5) self.collected_attn_logits = top_vals.detach().cpu().tolist() def forward( self, hidden_states, position_embeddings, attention_mask, position_ids, past_key_value, output_attentions = False, use_cache = False, cache_position = None, **kwargs, ): bsz, target_len, _ = hidden_states.size() q_len = target_len query_states = self.q_proj(hidden_states) key_states = self.k_proj(hidden_states) value_states = self.v_proj(hidden_states) query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2) key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2) value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2) cos, sin = position_embeddings query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin) if past_key_value is not None: # sin and cos are specific to RoPE models; cache_position needed for the static cache cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position} key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs) key_states = repeat_kv(key_states, self.num_key_value_groups) value_states = repeat_kv(value_states, self.num_key_value_groups) value_states = torch.cat(torch.chunk(value_states, 2, dim=1), dim=-1) value_states = value_states.repeat(1, 2, 1, 1) attn_weights = torch.matmul(query_states, key_states.transpose(2, 3)) / math.sqrt(self.head_dim) if attention_mask is not None: # no matter the length, we just slice it causal_mask = attention_mask[:, :, :, : key_states.shape[-2]] attn_weights = attn_weights + causal_mask # upcast attention to fp32 attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query_states.dtype) attn_weights = nn.functional.dropout(attn_weights, p=self.attention_dropout, training=self.training) lambda_full = self.lambda_init attn_output = torch.matmul(attn_weights, value_states) attn_output1, attn_output2 = torch.chunk(attn_output, 2, dim=1) attn_output = attn_output1 - lambda_full * attn_output2 attn_output = (1 - self.lambda_init) * self.groupnorm(attn_output) attn_output = attn_output.transpose(1, 2).contiguous() attn_output = attn_output.reshape(bsz, q_len, -1) attn_output = self.o_proj(attn_output) if not output_attentions: attn_weights = None return attn_output, attn_weights class DiffLlamaSdpaAttention(DiffLlamaAttention): """ DiffLlama attention module using torch.nn.functional.scaled_dot_product_attention. This module inherits from `DiffLlamaAttention` as the weights of the module stays untouched. The only changes are on the forward pass to adapt to SDPA API. """ # Adapted from DiffLlamaAttention.forward def forward( self, hidden_states: torch.Tensor, position_embeddings: Tuple[torch.Tensor, torch.Tensor], attention_mask: Optional[torch.Tensor] = None, position_ids: Optional[torch.LongTensor] = None, past_key_value = None, output_attentions: bool = False, use_cache: bool = False, cache_position: Optional[torch.LongTensor] = None, **kwargs, ): if output_attentions: # TODO: Improve this warning with e.g. `model.config.attn_implementation = "manual"` once this is implemented. logger.warning_once( "DiffLlamaModel is using DiffLlamaSdpaAttention, but `torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True`. Falling back to the manual attention implementation, " 'but specifying the manual implementation will be required from Transformers version v5.0.0 onwards. This warning can be removed using the argument `attn_implementation="eager"` when loading the model.' ) return super().forward( hidden_states=hidden_states, attention_mask=attention_mask, position_ids=position_ids, past_key_value=past_key_value, output_attentions=output_attentions, use_cache=use_cache, cache_position=cache_position, position_embeddings=position_embeddings, ) bsz, q_len, _ = hidden_states.size() query_states = self.q_proj(hidden_states) key_states = self.k_proj(hidden_states) value_states = self.v_proj(hidden_states) query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2) key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2) value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2) cos, sin = position_embeddings query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin) if past_key_value is not None: # sin and cos are specific to RoPE models; cache_position needed for the static cache cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position} key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs) key_states = repeat_kv(key_states, self.num_key_value_groups) value_states = repeat_kv(value_states, self.num_key_value_groups) value_states = torch.cat(torch.chunk(value_states, 2, dim=1), dim=-1) value_states = value_states.repeat(1, 2, 1, 1) causal_mask = attention_mask if attention_mask is not None: causal_mask = causal_mask[:, :, :, : key_states.shape[-2]] # SDPA with memory-efficient backend is currently (torch==2.1.2) bugged with non-contiguous inputs with custom attn_mask, # Reference: https://github.com/pytorch/pytorch/issues/112577. if query_states.device.type == "cuda" and causal_mask is not None: query_states = query_states.contiguous() key_states = key_states.contiguous() value_states = value_states.contiguous() # We dispatch to SDPA's Flash Attention or Efficient kernels via this `is_causal` if statement instead of an inline conditional assignment # in SDPA to support both torch.compile's dynamic shapes and full graph options. An inline conditional prevents dynamic shapes from compiling. is_causal = True if causal_mask is None and q_len > 1 else False attn_output = torch.nn.functional.scaled_dot_product_attention( query_states, key_states, value_states, attn_mask=causal_mask, dropout_p=self.attention_dropout if self.training else 0.0, is_causal=is_causal, ) attn_output1, attn_output2 = torch.chunk(attn_output, 2, dim=1) lambda_full = self.lambda_init attn_output = attn_output1 - lambda_full * attn_output2 attn_output = (1 - self.lambda_init) * self.groupnorm(attn_output) attn_output = attn_output.transpose(1, 2).contiguous() attn_output = attn_output.view(bsz, q_len, -1) attn_output = self.o_proj(attn_output) return attn_output, None def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor: """ This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch, num_key_value_heads, seqlen, head_dim) to (batch, num_attention_heads, seqlen, head_dim) """ batch, num_key_value_heads, slen, head_dim = hidden_states.shape if n_rep == 1: return hidden_states hidden_states = hidden_states[:, :, None, :, :].expand(batch, num_key_value_heads, n_rep, slen, head_dim) return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim) def eager_attention_forward( module: nn.Module, query: torch.Tensor, key: torch.Tensor, value: torch.Tensor, attention_mask: Optional[torch.Tensor], scaling: float, dropout: float = 0.0, **kwargs, ): key_states = repeat_kv(key, module.num_key_value_groups) value_states = repeat_kv(value, module.num_key_value_groups) temperature = kwargs.get("temperature", 1.0) attn_weights = torch.matmul(query, key_states.transpose(2, 3)) * scaling / temperature if attention_mask is not None: causal_mask = attention_mask[:, :, :, : key_states.shape[-2]] attn_weights = attn_weights + causal_mask attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query.dtype) attn_weights = nn.functional.dropout(attn_weights, p=dropout, training=module.training) attn_output = torch.matmul(attn_weights, value_states) attn_output = attn_output.transpose(1, 2).contiguous() return attn_output, attn_weights class LlamaRMSNorm(nn.Module): def __init__(self, hidden_size, eps=1e-6): """ LlamaRMSNorm is equivalent to T5LayerNorm """ super().__init__() self.weight = nn.Parameter(torch.ones(hidden_size)) self.variance_epsilon = eps def forward(self, hidden_states): input_dtype = hidden_states.dtype hidden_states = hidden_states.to(torch.float32) variance = hidden_states.pow(2).mean(-1, keepdim=True) hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon) return self.weight * hidden_states.to(input_dtype) def extra_repr(self): return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}" class LlamaRotaryEmbedding(nn.Module): def __init__(self, config: LlamaConfig, device=None): super().__init__() # BC: "rope_type" was originally "type" if hasattr(config, "rope_scaling") and config.rope_scaling is not None: self.rope_type = config.rope_scaling.get("rope_type", config.rope_scaling.get("type")) else: self.rope_type = "default" self.max_seq_len_cached = config.max_position_embeddings self.original_max_seq_len = config.max_position_embeddings self.config = config self.rope_init_fn = ROPE_INIT_FUNCTIONS[self.rope_type] inv_freq, self.attention_scaling = self.rope_init_fn(self.config, device) self.register_buffer("inv_freq", inv_freq, persistent=False) self.original_inv_freq = self.inv_freq def _dynamic_frequency_update(self, position_ids, device): """ dynamic RoPE layers should recompute `inv_freq` in the following situations: 1 - growing beyond the cached sequence length (allow scaling) 2 - the current sequence length is in the original scale (avoid losing precision with small sequences) """ seq_len = torch.max(position_ids) + 1 if seq_len > self.max_seq_len_cached: # growth inv_freq, self.attention_scaling = self.rope_init_fn(self.config, device, seq_len=seq_len) self.register_buffer("inv_freq", inv_freq, persistent=False) # TODO joao: may break with compilation self.max_seq_len_cached = seq_len if seq_len < self.original_max_seq_len and self.max_seq_len_cached > self.original_max_seq_len: # reset # This .to() is needed if the model has been moved to a device after being initialized (because # the buffer is automatically moved, but not the original copy) self.original_inv_freq = self.original_inv_freq.to(device) self.register_buffer("inv_freq", self.original_inv_freq, persistent=False) self.max_seq_len_cached = self.original_max_seq_len @torch.no_grad() def forward(self, x, position_ids): if "dynamic" in self.rope_type: self._dynamic_frequency_update(position_ids, device=x.device) # Core RoPE block inv_freq_expanded = self.inv_freq[None, :, None].float().expand(position_ids.shape[0], -1, 1) position_ids_expanded = position_ids[:, None, :].float() # Force float32 (see https://github.com/huggingface/transformers/pull/29285) device_type = x.device.type device_type = device_type if isinstance(device_type, str) and device_type != "mps" else "cpu" with torch.autocast(device_type=device_type, enabled=False): freqs = (inv_freq_expanded.float() @ position_ids_expanded.float()).transpose(1, 2) emb = torch.cat((freqs, freqs), dim=-1) cos = emb.cos() sin = emb.sin() # Advanced RoPE types (e.g. yarn) apply a post-processing scaling factor, equivalent to scaling attention cos = cos * self.attention_scaling sin = sin * self.attention_scaling return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype) def rotate_half(x): """Rotates half the hidden dims of the input.""" x1 = x[..., : x.shape[-1] // 2] x2 = x[..., x.shape[-1] // 2 :] return torch.cat((-x2, x1), dim=-1) def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1): """Applies Rotary Position Embedding to the query and key tensors. Args: q (`torch.Tensor`): The query tensor. k (`torch.Tensor`): The key tensor. cos (`torch.Tensor`): The cosine part of the rotary embedding. sin (`torch.Tensor`): The sine part of the rotary embedding. position_ids (`torch.Tensor`, *optional*): Deprecated and unused. unsqueeze_dim (`int`, *optional*, defaults to 1): The 'unsqueeze_dim' argument specifies the dimension along which to unsqueeze cos[position_ids] and sin[position_ids] so that they can be properly broadcasted to the dimensions of q and k. For example, note that cos[position_ids] and sin[position_ids] have the shape [batch_size, seq_len, head_dim]. Then, if q and k have the shape [batch_size, heads, seq_len, head_dim], then setting unsqueeze_dim=1 makes cos[position_ids] and sin[position_ids] broadcastable to the shapes of q and k. Similarly, if q and k have the shape [batch_size, seq_len, heads, head_dim], then set unsqueeze_dim=2. Returns: `tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding. """ cos = cos.unsqueeze(unsqueeze_dim) sin = sin.unsqueeze(unsqueeze_dim) q_embed = (q * cos) + (rotate_half(q) * sin) k_embed = (k * cos) + (rotate_half(k) * sin) return q_embed, k_embed class LlamaMLP(nn.Module): def __init__(self, config): super().__init__() self.config = config self.hidden_size = config.hidden_size self.intermediate_size = config.intermediate_size self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=config.mlp_bias) self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=config.mlp_bias) self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=config.mlp_bias) self.act_fn = ACT2FN[config.hidden_act] def forward(self, x): down_proj = self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x)) return down_proj