modeling_i3.py

# modeling_i3.py
from transformers import PreTrainedModel
from transformers.modeling_outputs import CausalLMOutputWithPast
from configuration_i3 import I3Config
from i3_architecture import i3Model  # your actual i3 implementation

class I3ForCausalLM(PreTrainedModel):
    config_class = I3Config

    def __init__(self, config):
        super().__init__(config)
        self.model = i3Model(
            vocab_size=config.vocab_size,
            d_model=config.d_model,
            n_layers=config.n_layers,
            n_heads=config.n_heads,
            max_seq_len=config.max_seq_len,
            rank=config.rank,
            d_state=config.d_state,
        )
        self.lm_head = torch.nn.Linear(config.d_model, config.vocab_size, bias=False)
        self.post_init()

    def forward(self, input_ids, labels=None, attention_mask=None, **kwargs):
        outputs = self.model(input_ids)
        logits = self.lm_head(outputs)
        
        loss = None
        if labels is not None:
            # Shift so that tokens < n predict n
            shift_logits = logits[..., :-1, :].contiguous()
            shift_labels = labels[..., 1:].contiguous()
            loss_fct = torch.nn.CrossEntropyLoss()
            loss = loss_fct(shift_logits.view(-1, shift_logits.size(-1)), 
                           shift_labels.view(-1))

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

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