Support Qwen 3 <think> blocks and add 14B model

index 6cebdce..1d17ef7 100644 --- a/.github/workflows/ci.yml +++ b/.github/workflows/ci.yml @@ -4,7 +4,10 @@ name: CI # Splits formatting, clippy, and tests into separate jobs for clearer reporting # across supported desktop targets. -on: push +on: + push: + pull_request: + workflow_dispatch: env: CARGO_TERM_COLOR: always

index 5148b6b..108c092 100644 --- a/src/chat.rs +++ b/src/chat.rs @@ -30,6 +30,44 @@ use ratatui::{ use tokio::sync::mpsc; use tokio::time::{Duration, Instant, interval}; +// ── Think-block stripping ───────────────────────────────────────────────────── + +/// Strip `<think>…</think>` blocks from a model response. +/// +/// Qwen 3 models emit `<think>…</think>` before the real answer. This +/// function separates the thinking content from the visible reply so the +/// UI can render them differently (dimmed / collapsed). +/// +/// Returns `(thinking_text, visible_reply)`. Either may be empty. +pub(crate) fn strip_think_blocks(raw: &str) -> (String, String) { + let mut thinking = String::new(); + let mut remainder = raw; + + while let Some(start) = remainder.find("<think>") { + // Text before <think> is visible. + let before = &remainder[..start]; + if let Some(end) = remainder[start..].find("</think>") { + let block = &remainder[start + 7..start + end]; + thinking.push_str(block.trim()); + remainder = &remainder[start + end + 8..]; + // Prepend any text before <think> to the leftover. + if !before.trim().is_empty() { + // Unusual — text before <think>. Keep it visible. + let mut combined = before.to_string(); + combined.push_str(remainder); + return (thinking, combined.trim().to_string()); + } + } else { + // Unclosed <think> — treat rest as thinking (model ran out of tokens). + thinking.push_str(remainder[start + 7..].trim()); + remainder = before; + break; + } + } + + (thinking, remainder.trim().to_string()) +} + // ── Message types ───────────────────────────────────────────────────────────── #[derive(Clone, Copy, PartialEq, Eq)] @@ -44,6 +82,8 @@ enum Role { struct ChatMessage { role: Role, text: String, + /// Extracted `<think>…</think>` content, if any (Qwen 3 reasoning). + think_block: Option<String>, } impl ChatMessage { @@ -51,13 +91,17 @@ impl ChatMessage { Self { role: Role::User, text: text.into(), + think_block: None, } } fn assistant(text: impl Into<String>) -> Self { + let raw = text.into(); + let (think, visible) = strip_think_blocks(&raw); Self { role: Role::Assistant, - text: text.into(), + text: visible, + think_block: if think.is_empty() { None } else { Some(think) }, } } @@ -65,6 +109,7 @@ impl ChatMessage { Self { role: Role::System, text: text.into(), + think_block: None, } } @@ -72,6 +117,7 @@ impl ChatMessage { Self { role: Role::Banner, text: text.into(), + think_block: None, } } } @@ -844,6 +890,24 @@ fn render_chat_message<'a>(lines: &mut Vec<Line<'a>>, msg: &ChatMessage) { } } Role::Assistant => { + // Show thinking block dimmed if present. + if let Some(ref think) = msg.think_block { + let think_summary = if think.len() > 120 { + format!("{}…", &think[..120]) + } else { + think.clone() + }; + lines.push(Line::from(vec![ + Span::styled("💭 ", Style::default().fg(Color::DarkGray)), + Span::styled( + think_summary, + Style::default() + .fg(Color::DarkGray) + .add_modifier(Modifier::ITALIC), + ), + ])); + } + for (i, segment) in text_lines.iter().enumerate() { let mut spans = Vec::new(); if i == 0 { @@ -1509,3 +1573,35 @@ async fn event_loop<B: ratatui::backend::Backend>( Ok(()) } + +#[cfg(test)] +mod tests { + use super::strip_think_blocks; + + #[test] + fn strip_think_blocks_separates_thinking_and_visible_reply() { + let raw = "<think>I should inspect the code first.</think>Here is the fix."; + let (thinking, visible) = strip_think_blocks(raw); + + assert_eq!(thinking, "I should inspect the code first."); + assert_eq!(visible, "Here is the fix."); + } + + #[test] + fn strip_think_blocks_handles_unclosed_think_block() { + let raw = "<think>I am still reasoning about the bug"; + let (thinking, visible) = strip_think_blocks(raw); + + assert_eq!(thinking, "I am still reasoning about the bug"); + assert_eq!(visible, ""); + } + + #[test] + fn strip_think_blocks_leaves_plain_text_untouched() { + let raw = "No hidden reasoning here."; + let (thinking, visible) = strip_think_blocks(raw); + + assert_eq!(thinking, ""); + assert_eq!(visible, "No hidden reasoning here."); + } +}

index c8b0db7..83f014c 100644 --- a/src/main.rs +++ b/src/main.rs @@ -1120,7 +1120,10 @@ impl Agent for SiGitAgent { String::new() } } else { - reply_text + // Strip Qwen 3 `<think>…</think>` blocks — the editor doesn't + // need to see internal reasoning tokens. + let (_think, visible) = chat::strip_think_blocks(&reply_text); + visible }; if !final_text.is_empty() {

index 5713a17..3f4d4c7 100644 --- a/src/models.rs +++ b/src/models.rs @@ -44,6 +44,7 @@ pub(crate) fn model_id_to_config(model_id: &str) -> Option<GgufModelConfig> { Some(match model_id { "bartowski/Qwen_Qwen3-4B-GGUF" => GgufModelConfig::qwen3_4b(), "bartowski/Qwen_Qwen3-8B-GGUF" => GgufModelConfig::qwen3_8b(), + "bartowski/Qwen_Qwen3-14B-GGUF" => GgufModelConfig::qwen3_14b(), "bartowski/Qwen_Qwen3-1.7B-GGUF" => GgufModelConfig::qwen3_1_7b(), "bartowski/Qwen2.5-3B-Instruct-GGUF" => GgufModelConfig::qwen25_3b(), "bartowski/Qwen2.5-1.5B-Instruct-GGUF" => GgufModelConfig::qwen25_1_5b(), @@ -61,6 +62,7 @@ fn is_tool_calling(model_id: &str) -> bool { model_id, "bartowski/Qwen_Qwen3-4B-GGUF" | "bartowski/Qwen_Qwen3-8B-GGUF" + | "bartowski/Qwen_Qwen3-14B-GGUF" | "bartowski/Qwen_Qwen3-1.7B-GGUF" | "bartowski/Qwen2.5-Coder-7B-Instruct-GGUF" )