Need help with messy outputs

[MFZORO]

So I've managed to at least get far enough for the endpoint to work. How do I configure it to just output raw translations, just like GoogleTranslate, DeepL and the other endpoints. Also where is the correct place to 'answer' the questions that it's trying to communicate to me?

[MFZORO]

`apiKey: "None"
apiKeyRequired: false
url: "http://localhost:11434/api/chat"
model: "microai/suzume-llama3"
modelParams:
temperature: 0.2
max_tokens: 4096
top_p: 0.9
top_k: 40
min_p: 0.5
frequency_penalty: 0
presence_penalty: 0
num_ctx: 8192

systemPrompt: |
Your task is to only translate Japanese text into English without any additional commentary, explanations, or extraneous information.
Ignore English text or content that is already in English.

Translation Guidelines:

• Directly translate Japanese into English.
• Do not add any extra text, descriptions, or context.
• If the text is already in English, do not translate it.
• Avoid any form of explanation, commentary, or analysis.
• Do not include HTML, markdown, or special formatting.

Gender-Neutral Language:

• Default to gender-neutral pronouns ("they/them"), unless specified otherwise by the text.

Names:

• Translate Japanese names into their Romanized form (Romaji) when applicable.
• Do not translate English names.

Cultural Sensitivity:

• If needed, use culturally relevant terms but avoid additional explanations of those terms.

Clarity and Accuracy:

• Ensure the translated text is grammatically correct and clear, while maintaining the original meaning.
• Do not adjust tone or style, except for making sure the text is fluent and natural in English.

Do not include any explanations or unnecessary details.`

Current prompt file config. It's still trying to translate english etc....

It's also extremely slow.

[joshfreitas1984]

You will want to tweak the prompt to your model. What I generally do is first prompt a new model with:

Optimise the system prompt for this model: <Paste system prompt> then use it.

My guess is the Phi4 Prompt is causing issues with your llama3 model

[MFZORO]

You will want to tweak the prompt to your model. What I generally do is first prompt a new model with:

Optimise the system prompt for this model: <Paste system prompt> then use it.

My guess is the Phi4 Prompt is causing issues with your llama3 model

Gave it a shot but it seems to have not changed much. Is there a solid japanese > english model that you'd recommend?

It also seems to ignore my prompts telling it not to translate english text