char-fork/llama.cpp

git clone https://git.t4t.associates/char-fork/llama.cpp

Andy Williamschat : pass reasoning_effort to template7e4c0a968

char/aliceai-t5-moe
brancheschar/aliceai-t5-moeall refs →
1.0 KiB40 linesraw
1#pragma once
2
3#include "runtime.h"
4
5#include <string>
6#include <map>
7
8namespace jinja {
9
10struct caps {
11    bool supports_tools = true;
12    bool supports_tool_calls = true;
13    bool supports_system_role = true;
14    bool supports_parallel_tool_calls = true;
15
16    // supports preserve reasoning trace in the full history, not just the last assistant message
17    bool supports_preserve_reasoning = false;
18
19    // supports reasoning effort levels
20    bool supports_reasoning_effort = false;
21
22    // one of the 2 content capabilities must be true
23    bool supports_string_content = true;
24    bool supports_typed_content = false;
25
26    bool supports_object_arguments = false;
27
28    // for reporting on server
29    std::map<std::string, bool> to_map() const;
30
31    // for debugging
32    std::string to_string() const;
33};
34
35caps caps_get(jinja::program & prog);
36
37void caps_apply_preserve_reasoning(jinja::context & ctx, bool enabled);
38void caps_apply_reasoning_effort(jinja::context & ctx, const std::string & effort);
39
40} // namespace jinja