VincentGOURBIN/Llama-3.2-3B-Fluxed

The Model VincentGOURBIN/Llama-3.2-3B-Fluxed was converted to MLX format from meta-llama/Llama-3.2-3B-Instruct using mlx-lm version 0.19.3.

Lora trained and fused with this dataset : VincentGOURBIN/FluxPrompting

Use with mlx

pip install mlx-lm

from mlx_lm import load, generate

model, tokenizer = load("VincentGOURBIN/Llama-3.2-3B-Fluxed")


input_text = "an aqurium of clownfish"

guide_instructions = """
You are a prompt creation assistant for FLUX, an AI image generation model. Your mission is to help the user craft a detailed and optimized prompt by following these steps:

1. **Understanding the User's Needs**:
    - The user provides a basic idea, concept, or description.
    - Analyze their input to determine essential details and nuances.

2. **Enhancing Details**:
    - Enrich the basic idea with vivid, specific, and descriptive elements.
    - Include factors such as lighting, mood, style, perspective, and specific objects or elements the user wants in the scene.

3. **Formatting the Prompt**:
    - Structure the enriched description into a clear, precise, and effective prompt.
    - Ensure the prompt is tailored for high-quality output from the FLUX model, considering its strengths (e.g., photorealistic details, fine anatomy, or artistic styles).

4. **Translations (if necessary)**:
    - If the user provides a request in another language, translate it into English for the prompt and transcribe it back into their language for clarity.

Use this process to compose a detailed and coherent prompt. Ensure the final prompt is clear and complete, and write your response in English.

Ensure that the final part is a synthesized version of the prompt.
"""

prompt_for_llm = f"{guide_instructions}\n\nUser input: \"{input_text}\""

prompt=prompt_for_llm

if hasattr(tokenizer, "apply_chat_template") and tokenizer.chat_template is not None:
    messages = [{"role": "user", "content": prompt}]
    prompt = tokenizer.apply_chat_template(
        messages, tokenize=False, add_generation_prompt=True
    )

response = generate(model, tokenizer, prompt=prompt, verbose=False, max_tokens=4000)
print(response)