llama_diffusion_model.py

import torch
import torch.nn as nn
from torch.amp import autocast
from transformers import AutoModelForCausalLM, PreTrainedModel, PretrainedConfig
from transformers.models.llama.modeling_llama import LlamaAttention
from peft import LoraConfig, get_peft_model
import os
from typing import Optional, Tuple

hf_token = os.getenv("HF_TOKEN")

class BidirectionalLlamaAttention(LlamaAttention):
    def __init__(self, original_layer, masking='unidirectional'):
        super().__init__(original_layer.config, layer_idx=original_layer.layer_idx)
        self.masking = masking
        self.q_proj.weight = original_layer.q_proj.weight
        self.k_proj.weight = original_layer.k_proj.weight
        self.v_proj.weight = original_layer.v_proj.weight
        self.o_proj.weight = original_layer.o_proj.weight

    def repeat_kv(self, hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
        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(self, module: nn.Module, query, key, value, attention_mask, scaling, dropout=0.0, **kwargs):
        key_states = self.repeat_kv(key, module.num_key_value_groups)
        value_states = self.repeat_kv(value, module.num_key_value_groups)
        attn_weights = torch.matmul(query, key_states.transpose(2, 3)) * scaling

        if attention_mask is not None:
            attn_mask = (1.0 - attention_mask) * float('-inf')
            attn_mask = attn_mask.to(dtype=query.dtype)
            attn_weights = attn_weights + attn_mask

        attn_weights = nn.functional.softmax(attn_weights, dim=-1).to(query.dtype)
        attn_weights = nn.functional.dropout(attn_weights, p=dropout, training=module.training)
        attn_output = torch.matmul(attn_weights, value_states).transpose(1, 2).contiguous()
        return attn_output, attn_weights

    def rotate_half(self, x):
        x1 = x[..., : x.shape[-1] // 2]
        x2 = x[..., x.shape[-1] // 2:]
        return torch.cat((-x2, x1), dim=-1)

    def apply_rotary_pos_emb(self, q, k, cos, sin, unsqueeze_dim=1):
        cos = cos.unsqueeze(unsqueeze_dim)
        sin = sin.unsqueeze(unsqueeze_dim)
        q_embed = (q * cos) + (self.rotate_half(q) * sin)
        k_embed = (k * cos) + (self.rotate_half(k) * sin)
        return q_embed, k_embed

    def forward(self, hidden_states, position_embeddings, attention_mask=None, past_key_value=None, cache_position=None, **kwargs):
        input_shape = hidden_states.shape[:-1]
        hidden_shape = (*input_shape, -1, self.head_dim)

        query_states = self.q_proj(hidden_states).view(hidden_shape).transpose(1, 2)
        key_states = self.k_proj(hidden_states).view(hidden_shape).transpose(1, 2)
        value_states = self.v_proj(hidden_states).view(hidden_shape).transpose(1, 2)

        cos, sin = position_embeddings
        query_states, key_states = self.apply_rotary_pos_emb(query_states, key_states, cos, sin)

        if past_key_value is not None:
            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)

        attn_output, attn_weights = self.eager_attention_forward(
            self, query_states, key_states, value_states, attention_mask,
            dropout=0.0 if not self.training else self.attention_dropout, scaling=self.scaling, **kwargs
        )

        attn_output = attn_output.reshape(*input_shape, -1).contiguous()
        return self.o_proj(attn_output), attn_weights

class CustomTransformerConfig(PretrainedConfig):
    def __init__(self, vocab_size=128256, hidden_size=4096, num_layers=32, num_heads=32, prediction_chunk=256, dropout=0,
                 max_position_embeddings=4096, masking_type="bidirectional", **kwargs):
        super().__init__(**kwargs)
        self.vocab_size = vocab_size
        self.hidden_size = hidden_size
        self.num_layers = num_layers
        self.num_heads = num_heads
        self.dropout = dropout
        self.prediction_chunk = prediction_chunk
        self.max_position_embeddings = max_position_embeddings
        self.input_size = prediction_chunk
        self.masking_type = masking_type

class CustomTransformerModel(PreTrainedModel):
    config_class = CustomTransformerConfig

    def __init__(self, config):
        super().__init__(config)
        self.llama = AutoModelForCausalLM.from_pretrained("meta-llama/Llama-3.2-3B", torch_dtype=torch.float16, device_map="auto", token=hf_token)
        self.llama.resize_token_embeddings(config.vocab_size)

        # for i, layer in enumerate(self.llama.model.layers):
        #     layer.self_attn = BidirectionalLlamaAttention(layer.self_attn, masking=config.masking_type)

        for param in self.llama.parameters():
            param.requires_grad = False
        for param in self.llama.lm_head.parameters():
            param.requires_grad = True

        lora_config = LoraConfig(
            r=512, lora_alpha=512, lora_dropout=0.0,
            target_modules=["q_proj", "v_proj", "k_proj", "o_proj"],
            bias="none", task_type=None
        )

        self.llama = get_peft_model(self.llama, lora_config)
        self.llama.print_trainable_parameters()
        # self.llama = self.llama.to(torch.float16)

    def forward(self, input_ids, labels=None, **kwargs):
        batch_size, seq_len = input_ids.shape
        assert seq_len == self.config.prediction_chunk, f"Expected input length {self.config.prediction_chunk}, got {seq_len}"

        # Build attention mask
        device = input_ids.device

        masking_type = getattr(self.config, "masking_type", "bidirectional")
        if masking_type == 'bidirectional':
            base_mask = torch.ones(seq_len, seq_len, dtype=torch.bool, device=device)
        elif masking_type == 'bidirectional_masked':
            base_mask = torch.ones(seq_len, seq_len, dtype=torch.bool, device=device)
            base_mask.fill_diagonal_(False)
        elif masking_type == 'unidirectional':
            base_mask = torch.tril(torch.ones(seq_len, seq_len, dtype=torch.bool, device=device))
        else:
            raise ValueError(f"Unknown masking type: {self.config.masking_type}")

        attention_mask = base_mask.unsqueeze(0).unsqueeze(1).expand(batch_size, 1, seq_len, seq_len).clone()
        attention_mask = attention_mask.to(dtype=torch.float32)  # required for SDPA and Flash attention


        with autocast("cuda", dtype=torch.float16):
            outputs = self.llama(
                input_ids,
                attention_mask=attention_mask,
                output_hidden_states=True,
                use_cache=False,
                **kwargs
            )

        logits = outputs.logits[:, :, :self.config.vocab_size].view(batch_size, seq_len, self.config.vocab_size)

        loss = None
        if labels is not None:
            assert labels.shape == (batch_size, seq_len), f"Labels shape mismatch: expected ({batch_size}, {seq_len}), got {labels.shape}"
            loss_fct = nn.CrossEntropyLoss()
            loss = loss_fct(logits.view(-1, self.config.vocab_size), labels.view(-1))

        return {"loss": loss, "logits": logits} if loss is not None else {"logits": logits}

def disable_dropout(model):
    for name, module in model.named_modules():
        if isinstance(module, nn.Dropout):
            setattr(model, name, nn.Identity())
    return model