rlx-gemma 0.2.11

Gemma / Gemma 2 causal LMs for RLX
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69

// RLX — versatile ML compiler + runtime.
// Copyright (C) 2026 Eugene Hauptmann, Nataliya Kosmyna.
//
// This program is free software: you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation, version 3.
//
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU General Public License
// along with this program. If not, see <https://www.gnu.org/licenses/>.

//! Chat-template prompt encoding for Gemma GGUF / HF checkpoints.

use anyhow::Result;
use rlx_cli::{ChatMessage, auto_chat_template};
use rlx_qwen35::{decode_ids_auto, encode_prompt_auto};
use std::path::Path;

/// Render the embedded GGUF chat template (when present) and tokenize.
///
/// Falls back to raw `user` text when no template is embedded or rendering
/// fails. Gemma 4 thinking models (e.g. the Composer/Fable coder fine-tune)
/// rely on the template to open the thought channel before generation.
pub fn encode_chat_prompt_auto(
    weights: &Path,
    tokenizer: Option<&Path>,
    system: Option<&str>,
    user: &str,
    use_chat_template: bool,
) -> Result<Vec<u32>> {
    let text = if use_chat_template {
        match auto_chat_template(weights) {
            Ok(tmpl) => {
                let mut msgs = Vec::new();
                if let Some(s) = system {
                    msgs.push(ChatMessage::system(s.to_string()));
                }
                msgs.push(ChatMessage::user(user.to_string()));
                tmpl.render(&msgs, true)?
            }
            Err(_) => user.to_string(),
        }
    } else {
        user.to_string()
    };
    let mut ids = if use_chat_template && text.starts_with("<bos>") {
        let body = text.strip_prefix("<bos>").unwrap_or(&text);
        let mut v = encode_prompt_auto(weights, tokenizer, body)?;
        if v.first() != Some(&2) {
            v.insert(0, 2);
        }
        v
    } else {
        encode_prompt_auto(weights, tokenizer, &text)?
    };
    if ids.len() >= 2 && ids[0] == 107 && ids[1] == 2 {
        ids.remove(0);
    }
    Ok(ids)
}

/// Decode one generated token id to a UTF-8 fragment (best-effort streaming).
pub fn decode_token_auto(weights: &Path, tokenizer: Option<&Path>, tok: u32) -> Result<String> {
    decode_ids_auto(weights, tokenizer, &[tok], false)
}