llama_edit

#!/bin/bash

# start llama.cpp service
server="http://127.0.0.1:8087/v1/chat/completions"
model="Bonsai-4B-Q1_0"
systemctl --user start llama.service

system_message='You are a text correction assistant for live streaming. Your only job is to correct grammar, spelling, punctuation, and clarity in user-submitted text. Translate to English, if necessary.

RULES:
1. Apply all necessary corrections (grammar, spelling, punctuation, word choice, clarity).
2. If the text contains obvious errors or awkward phrasing, correct them silently—do not explain or ask questions.
3. If no corrections are needed, return the text exactly as provided.
4. Output ONLY the corrected, English text in markdown format.
5. Do not join the conversation, respond to content, ask clarifying questions, or add commentary.
6. Do not interrupt or acknowledge the correction process.
7. Simply echo back the cleaned text.

Example:
- Input: "What is the capital of France, or rather England?"
- Output: "What is the capital of England?"

Input text to correct:'

# Build the JSON payload
payload=$(jq -n \
  --arg system "$system_message" \
  --arg user "Reply with corrected text: $*" \
  --arg model "$model" \
  '{
    system: $system,
    messages: [
{role: "user", content: "Reply with corrected text: This is a message for Bobarino, I mean Steve. Tell him to bring a saw to the job site."},
{role: "assistant", content: "Hi Steve,\n\nPlease bring two saws to the job site."},
{role: "user", content: "Reply with corrected text: It'\''s good to see you, um, great, um... Can you meet me at 8, I mean 9 PM?"},
{role: "assistant", content: "It'\''s good to see you. Can you meet me at 9 PM?"},
{role: "user", content: "Reply with corrected text: What is the capital of France, or rather England?"},
{role: "assistant", content: "What is the capital of England?"},
{role: "user", content: "Reply with corrected text: In bold text, um, we'\''re going to put a heading that says, Let'\''s go."},
{role: "assistant", content: "**Let'\''s go**"},
{role: "user", content: "Reply with corrected text: Under that, let'\''s make a list. Ok let'\''s put saws, brake pads, sandpaper, I mean sanding stones, and some oil"},
{role: "assistant", content: "\n - Saws\n - Brake pads\n - Sanding stones\n - Oil"},
{role: "user", content: "Reply with corrected text: Greetings, my man!"},
{role: "assistant", content: "Greetings, my man!"},
{role: "user", content: "Reply with corrected text: Can you feel the love tonight?"},
{role: "assistant", content: "Can you feel the love tonight?"},
{role: "user", content: $user}
    ],
    n_predict: 512,
    temperature: 0.1,
    stream: false,
    model: $model
  }')

# Echo the query for debugging
#echo "=== SYSTEM MESSAGE ==="
#echo "$system_message"
#echo ""
#echo "=== USER QUERY ==="
#echo "$*"
#echo ""
#echo "=== JSON PAYLOAD ==="
#echo "$payload" | jq .
#echo ""

# Send request to llama-server
response=$(curl -X POST "$server" \
  -H "Content-Type: application/json" \
  -d "$payload" \
  -s | jq -r '.choices[0].message.content')

echo "$response"