PrepareBuddy IELTS-2B — MLX (Apple Silicon)

The MLX 4-bit build of PrepareBuddy IELTS-2B, for fast local use on Apple Silicon Macs — including LM Studio. Generates IELTS Academic practice content across all four sections.

Content generator, not an assessment tool. A fine-tune of Qwen3.5-2B — not a foundation model. Full details, examples, the 2B/4B/9B comparison, and the grounding+verification how-to are on the main page: 👉 https://huggingface.co/preparebuddy/ielts-2b

Run in LM Studio (Mac)

  1. Search preparebuddy/ielts-2b-mlx and download it (or it appears automatically if placed in your models folder).
  2. System prompt:

    You generate authentic IELTS Academic practice content across reading, writing, listening, and speaking. Produce passages, transcripts, tasks, questions, and answer keys or model answers as appropriate to the section. Use IELTS-style register: academic, neutral, factually plausible. This is content generation, not assessment.

  3. Prompt with the tag format, e.g.:
    <TEST=IELTS><SECTION=WRITING><TYPE=TASK2><DIFF=medium><TOPIC=social media> Generate an IELTS Writing Task 2 question with word limit and timing.
    
  4. Settings: temperature 0.3 for verdicts (TFNG/YNNG/MCQ), 0.7 for writing/speaking; top_p 0.9. Max tokens ~400.

Run with mlx-lm (CLI)

pip install mlx-lm
python -m mlx_lm generate --model preparebuddy/ielts-2b-mlx \
  --prompt "<TEST=IELTS><SECTION=SPEAKING><TYPE=PART2><DIFF=medium><TOPIC=a memorable journey> Generate an IELTS Speaking Part 2 cue card with a model answer." --temp 0.7

Tip: the 2B is the family's best verdict judge but weakest on facts — for reliable answer keys, generate against a real passage and use the re-checking loop (see the main page).

License

Apache-2.0, inheriting from Qwen3.5-2B. Built by PrepareBuddy.

Downloads last month
-
Safetensors
Model size
0.3B params
Tensor type
BF16
·
U32
·
MLX
Hardware compatibility
Log In to add your hardware

4-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for preparebuddy/ielts-2b-mlx

Finetuned
Qwen/Qwen3.5-2B
Quantized
(108)
this model