File /anaconda/envs/text2text-tagger/lib/python3.11/site-packages/unsloth/models/llama.py:218, in LlamaAttention_fast_forward_inference(self, hidden_states, past_key_value, position_ids, do_prefill, attention_mask)
    216     A = torch.matmul(A, Vnn, out = Qn)
    217 else:
--> 218     A = scaled_dot_product_attention(Qn, Knn, Vnn, attn_mask = attention_mask, is_causal = False)
    219 pass
    220 A = A.transpose(1, 2)

RuntimeError: The expanded size of the tensor (2047) must match the existing size (2956) at non-singleton dimension 3.  Target sizes: [2, 32, 1, 2047].  Tensor sizes: [2, 1, 1, 2956]

# Transformers scans dependencies in the modeling file, causing issues on conditional loading. The regex only ignores try/catch blocks, but not if statements
# if is_flash_attn_2_available():
_flash_supports_window_size = False
try:
    from flash_attn import flash_attn_func, flash_attn_varlen_func
    from flash_attn.bert_padding import index_first_axis, pad_input, unpad_input  # noqa

    _flash_supports_window_size = "window_size" in list(inspect.signature(flash_attn_func).parameters)
except ImportError as error:
    logger.warning(
        f"`flash-attention` package not found, consider installing for better performance: {error}."
    )
    if not _flash_supports_window_size:
        logger.warning(
            "Current `flash-attention` does not support `window_size`. Either upgrade or use `attn_implementation='eager'`."
        )

       use_sliding_windows = (
            _flash_supports_window_size
            and getattr(self.config, "sliding_window", None) is not None
            and kv_seq_len > self.config.sliding_window
        )