LFM2.5-CT-230M

This is an Audio-Text-to-Text Fusion Model combining:

• Audio Encoder: FastConformer (Parakeet TDT 0.6B-v3)
• Text Decoder: LiquidAI LFM2.5-230M

Fused using a soft latent projection interface as described in Click-to-Align (§3.2).

Loading with Transformers

You can load and run this model using Hugging Face Transformers with trust_remote_code=True:

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("ray0rf1re/lfm2.5-CT-230m")
model = AutoModelForCausalLM.from_pretrained(
    "ray0rf1re/lfm2.5-CT-230m",
    trust_remote_code=True,
    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32
)