LC-350M-smart (Lint Clean · course-correction)
On-device dictation cleanup with spoken self-repairs for Apple Silicon (MLX).
Fine-tuned from LiquidAI/LFM2.5-350M.
Example:
| Raw ASR | Clean |
|---|---|
| I wanna get the bag no not not bag my phone | I want to get my phone. |
| we should use Qwen wait no use Parakeet V3 | We should use Parakeet V3. |
Use LC-350M-light for conservative polish without aggressive rewrites.
Usage (MLX)
pip install mlx-lm
mlx_lm.generate --model vasanth009/LC-350M-smart \
--max-tokens 128 \
--prompt "Clean this voice dictation into what the speaker finally meant.\n\nI wanna get the bag no not not bag my phone"
Related
- Training repo: github.com/vasanthsreeram/lc-350m
- App: github.com/vasanthsreeram/macwispr
- Light variant: LC-350M-light
License
Inherits Liquid LFM base model terms. Training code in the GitHub repo is Apache-2.0.
- Downloads last month
- 151
Model size
55.4M params
Tensor type
F32
·
U32 ·
Hardware compatibility
Log In to add your hardware
4-bit
# Make sure mlx-lm is installed # pip install --upgrade mlx-lm # Generate text with mlx-lm from mlx_lm import load, generate model, tokenizer = load("vasanth009/LC-350M-smart") prompt = "Write a story about Einstein" messages = [{"role": "user", "content": prompt}] prompt = tokenizer.apply_chat_template( messages, add_generation_prompt=True ) text = generate(model, tokenizer, prompt=prompt, verbose=True)