modeling_quark.py

"""
Quark model implementation for HuggingFace Transformers.

Usage:
    from transformers import AutoModelForCausalLM, AutoTokenizer

    model = AutoModelForCausalLM.from_pretrained("ThingAI/Quark-135m-v0.2", trust_remote_code=True)
    tokenizer = AutoTokenizer.from_pretrained("ThingAI/Quark-135m-v0.2")

    inputs = tokenizer("Ciao, come stai?", return_tensors="pt")
    outputs = model.generate(**inputs, max_new_tokens=100, temperature=0.7, do_sample=True)
    print(tokenizer.decode(outputs[0], skip_special_tokens=True))
"""

import torch
import torch.nn as nn
import torch.nn.functional as F
from typing import Optional
from transformers import PreTrainedModel
from transformers.modeling_outputs import CausalLMOutputWithPast
from .configuration_quark import QuarkConfig


class QuarkRMSNorm(nn.Module):
    def __init__(self, dim: int, eps: float = 1e-5):
        super().__init__()
        self.eps = eps
        self.scale = nn.Parameter(torch.ones(dim))

    def forward(self, x):
        rms = x.float().pow(2).mean(-1, keepdim=True).add(self.eps).rsqrt()
        return (x.float() * rms).to(x.dtype) * self.scale


class QuarkRotaryEmbedding(nn.Module):
    def __init__(self, head_dim: int, max_seq_len: int, theta: float = 10000.0):
        super().__init__()
        assert head_dim % 2 == 0
        inv_freq = 1.0 / (theta ** (torch.arange(0, head_dim, 2).float() / head_dim))
        self.register_buffer("inv_freq", inv_freq, persistent=False)
        self._build_cache(max_seq_len)

    def _build_cache(self, seq_len: int):
        t = torch.arange(seq_len, device=self.inv_freq.device).float()
        freqs = torch.outer(t, self.inv_freq)
        emb = torch.cat([freqs, freqs], dim=-1)
        self.register_buffer("cos_cache", emb.cos()[None, None], persistent=False)
        self.register_buffer("sin_cache", emb.sin()[None, None], persistent=False)
        self._max_cached = seq_len

    @staticmethod
    def _rotate_half(x):
        x1, x2 = x.chunk(2, dim=-1)
        return torch.cat([-x2, x1], dim=-1)

    def forward(self, q, k):
        T = q.size(2)
        if T > self._max_cached:
            self._build_cache(T)
        cos = self.cos_cache[:, :, :T, :]
        sin = self.sin_cache[:, :, :T, :]
        q = q * cos + self._rotate_half(q) * sin
        k = k * cos + self._rotate_half(k) * sin
        return q, k


class QuarkAttention(nn.Module):
    """Grouped Query Attention (GQA)."""

    def __init__(self, config: QuarkConfig):
        super().__init__()
        self.n_heads = config.n_heads
        self.n_kv_heads = config.n_kv_heads
        self.n_groups = config.n_heads // config.n_kv_heads
        self.head_dim = config.head_dim

        self.q_proj = nn.Linear(config.d_model, config.n_heads * config.head_dim, bias=config.qkv_bias)
        self.k_proj = nn.Linear(config.d_model, config.n_kv_heads * config.head_dim, bias=config.qkv_bias)
        self.v_proj = nn.Linear(config.d_model, config.n_kv_heads * config.head_dim, bias=config.qkv_bias)
        self.o_proj = nn.Linear(config.n_heads * config.head_dim, config.d_model, bias=False)
        self.rope = QuarkRotaryEmbedding(config.head_dim, config.max_seq_len, config.rope_theta)

    def forward(self, x):
        B, T, _ = x.shape
        q = self.q_proj(x).view(B, T, self.n_heads, self.head_dim).transpose(1, 2)
        k = self.k_proj(x).view(B, T, self.n_kv_heads, self.head_dim).transpose(1, 2)
        v = self.v_proj(x).view(B, T, self.n_kv_heads, self.head_dim).transpose(1, 2)

        q, k = self.rope(q, k)

        if self.n_groups > 1:
            k = k.repeat_interleave(self.n_groups, dim=1)
            v = v.repeat_interleave(self.n_groups, dim=1)

        out = F.scaled_dot_product_attention(q, k, v, is_causal=True)
        out = out.transpose(1, 2).contiguous().view(B, T, self.n_heads * self.head_dim)
        return self.o_proj(out)


class QuarkFFN(nn.Module):
    """SwiGLU Feed-Forward Network."""

    def __init__(self, config: QuarkConfig):
        super().__init__()
        self.gate_proj = nn.Linear(config.d_model, config.d_ff, bias=False)
        self.up_proj = nn.Linear(config.d_model, config.d_ff, bias=False)
        self.down_proj = nn.Linear(config.d_ff, config.d_model, bias=False)

    def forward(self, x):
        return self.down_proj(F.silu(self.gate_proj(x)) * self.up_proj(x))


class QuarkBlock(nn.Module):
    """Transformer block with pre-norm."""

    def __init__(self, config: QuarkConfig):
        super().__init__()
        self.norm_attn = QuarkRMSNorm(config.d_model, config.rms_eps)
        self.attn = QuarkAttention(config)
        self.norm_ffn = QuarkRMSNorm(config.d_model, config.rms_eps)
        self.ffn = QuarkFFN(config)

    def forward(self, x):
        x = x + self.attn(self.norm_attn(x))
        x = x + self.ffn(self.norm_ffn(x))
        return x


class QuarkPreTrainedModel(PreTrainedModel):
    config_class = QuarkConfig
    base_model_prefix = "model"
    supports_gradient_checkpointing = False

    def _init_weights(self, module):
        std = 0.02
        if isinstance(module, nn.Linear):
            module.weight.data.normal_(mean=0.0, std=std)
            if module.bias is not None:
                module.bias.data.zero_()
        elif isinstance(module, nn.Embedding):
            module.weight.data.normal_(mean=0.0, std=std)


class QuarkForCausalLM(QuarkPreTrainedModel):
    """Quark model for causal language modeling."""

    def __init__(self, config: QuarkConfig):
        super().__init__(config)
        self.config = config

        self.embed_tokens = nn.Embedding(config.vocab_size, config.d_model)
        self.layers = nn.ModuleList([QuarkBlock(config) for _ in range(config.n_layers)])
        self.norm = QuarkRMSNorm(config.d_model, config.rms_eps)
        self.lm_head = nn.Linear(config.d_model, config.vocab_size, bias=False)

        # Weight tying
        self.lm_head.weight = self.embed_tokens.weight

        self.post_init()

    def get_input_embeddings(self):
        return self.embed_tokens

    def set_input_embeddings(self, value):
        self.embed_tokens = value

    def get_output_embeddings(self):
        return self.lm_head

    def set_output_embeddings(self, new_embeddings):
        self.lm_head = new_embeddings

    def forward(
        self,
        input_ids: torch.LongTensor,
        attention_mask: Optional[torch.Tensor] = None,
        labels: Optional[torch.LongTensor] = None,
        **kwargs,
    ) -> CausalLMOutputWithPast:
        h = self.embed_tokens(input_ids)

        for layer in self.layers:
            h = layer(h)

        h = self.norm(h)
        logits = self.lm_head(h)

        loss = None
        if labels is not None:
            shift_logits = logits[..., :-1, :].contiguous()
            shift_labels = labels[..., 1:].contiguous()
            loss = F.cross_entropy(
                shift_logits.view(-1, self.config.vocab_size),
                shift_labels.view(-1),
                ignore_index=-100,
            )

        return CausalLMOutputWithPast(
            loss=loss,
            logits=logits,
        )

    def prepare_inputs_for_generation(self, input_ids, **kwargs):
        return {"input_ids": input_ids}