selfware 0.6.1

Your personal AI workshop — software you own, software that lasts
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
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260

//! Quant catalog mirroring the Python `QUANT_CATALOG` dict.
//!
//! Keeping the entries in lock-step with `scripts/swebench_pro/run.py` so a user
//! invoking either path picks the same GGUF / alias / mmproj for a given label.

use once_cell::sync::Lazy;
use serde::{Deserialize, Serialize};
use std::collections::BTreeMap;

#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
pub enum BackendProfile {
    LlamaCpp,
    SGLang,
    VLLM,
}

#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
pub enum ThinkingPolicy {
    Enable,
    Disable,
    Preserve,
}

#[derive(Clone, Debug, Serialize, Deserialize)]
pub struct QuantSpec {
    pub label: String,
    /// GGUF filename (relative to ~/models/qwen36-quants).
    pub gguf: String,
    /// llama-server `--alias`.
    pub alias: String,
    /// mmproj filename (relative to ~/models/qwen36-quants).
    pub mmproj: String,
    /// Human-readable name.
    pub name: String,
    /// Recommended context size for this quant.
    pub ctx: u32,
    /// Maximum parallel slots for this quant (safety ceiling).
    pub max_parallel: u32,
    /// KV cache quant type: `"q4_0"`, `"q8_0"`, `"f16"`.
    pub kv_cache_type: String,
    /// `--tensor-split` value (e.g. `"24,24"` for 2x4090).
    pub tensor_split: Option<String>,
    /// Sampling temperature.
    pub temperature: f32,
    /// Whether thinking tokens are emitted.
    pub thinking_policy: ThinkingPolicy,
    /// Which backend engine serves this quant.
    pub backend: BackendProfile,
}

static CATALOG: Lazy<BTreeMap<String, QuantSpec>> = Lazy::new(|| {
    vec![
        QuantSpec {
            label: "Qwen3.6-27B-HauhauCS-IQ2_M".into(),
            gguf: "Qwen3.6-27B-Uncensored-HauhauCS-Aggressive-IQ2_M.gguf".into(),
            alias: "qwen3.6-27b-iq2m".into(),
            mmproj: "mmproj-Qwen3.6-27B-Uncensored-HauhauCS-Aggressive-f16.gguf".into(),
            name: "Qwen3.6-27B-HauhauCS-IQ2_M".into(),
            ctx: 262_144,
            max_parallel: 2,
            kv_cache_type: "q8_0".into(),
            tensor_split: Some("24,24".into()),
            temperature: 1.0,
            thinking_policy: ThinkingPolicy::Disable,
            backend: BackendProfile::LlamaCpp,
        },
        QuantSpec {
            label: "Qwen3.6-27B-HauhauCS-IQ3_XS".into(),
            gguf: "Qwen3.6-27B-Uncensored-HauhauCS-Aggressive-IQ3_XS.gguf".into(),
            alias: "qwen3.6-27b-iq3xs".into(),
            mmproj: "mmproj-Qwen3.6-27B-Uncensored-HauhauCS-Aggressive-f16.gguf".into(),
            name: "Qwen3.6-27B-HauhauCS-IQ3_XS".into(),
            ctx: 262_144,
            max_parallel: 2,
            kv_cache_type: "q8_0".into(),
            tensor_split: Some("24,24".into()),
            temperature: 1.0,
            thinking_policy: ThinkingPolicy::Disable,
            backend: BackendProfile::LlamaCpp,
        },
        QuantSpec {
            label: "Qwen3.6-27B-HauhauCS-IQ3_M".into(),
            gguf: "Qwen3.6-27B-Uncensored-HauhauCS-Aggressive-IQ3_M.gguf".into(),
            alias: "qwen3.6-27b-iq3m".into(),
            mmproj: "mmproj-Qwen3.6-27B-Uncensored-HauhauCS-Aggressive-f16.gguf".into(),
            name: "Qwen3.6-27B-HauhauCS-IQ3_M".into(),
            ctx: 262_144,
            max_parallel: 2,
            kv_cache_type: "q8_0".into(),
            tensor_split: Some("24,24".into()),
            temperature: 1.0,
            thinking_policy: ThinkingPolicy::Disable,
            backend: BackendProfile::LlamaCpp,
        },
        QuantSpec {
            label: "Qwen3.6-27B-HauhauCS-IQ4_XS".into(),
            gguf: "Qwen3.6-27B-Uncensored-HauhauCS-Aggressive-IQ4_XS.gguf".into(),
            alias: "qwen3.6-27b-iq4xs".into(),
            mmproj: "mmproj-Qwen3.6-27B-Uncensored-HauhauCS-Aggressive-f16.gguf".into(),
            name: "Qwen3.6-27B-HauhauCS-IQ4_XS".into(),
            ctx: 262_144,
            max_parallel: 2,
            kv_cache_type: "q8_0".into(),
            tensor_split: Some("24,24".into()),
            temperature: 1.0,
            thinking_policy: ThinkingPolicy::Disable,
            backend: BackendProfile::LlamaCpp,
        },
        QuantSpec {
            label: "Qwen3.6-27B-HauhauCS-Q2_K_P".into(),
            gguf: "Qwen3.6-27B-Uncensored-HauhauCS-Aggressive-Q2_K_P.gguf".into(),
            alias: "qwen3.6-27b-q2kp".into(),
            mmproj: "mmproj-Qwen3.6-27B-Uncensored-HauhauCS-Aggressive-f16.gguf".into(),
            name: "Qwen3.6-27B-HauhauCS-Q2_K_P".into(),
            ctx: 262_144,
            max_parallel: 2,
            kv_cache_type: "q8_0".into(),
            tensor_split: Some("24,24".into()),
            temperature: 1.0,
            thinking_policy: ThinkingPolicy::Disable,
            backend: BackendProfile::LlamaCpp,
        },
        QuantSpec {
            label: "Qwen3.6-27B-HauhauCS-Q3_K_P".into(),
            gguf: "Qwen3.6-27B-Uncensored-HauhauCS-Aggressive-Q3_K_P.gguf".into(),
            alias: "qwen3.6-27b-q3kp".into(),
            mmproj: "mmproj-Qwen3.6-27B-Uncensored-HauhauCS-Aggressive-f16.gguf".into(),
            name: "Qwen3.6-27B-HauhauCS-Q3_K_P".into(),
            ctx: 262_144,
            max_parallel: 2,
            kv_cache_type: "q8_0".into(),
            tensor_split: Some("24,24".into()),
            temperature: 1.0,
            thinking_policy: ThinkingPolicy::Disable,
            backend: BackendProfile::LlamaCpp,
        },
        QuantSpec {
            label: "Qwen3.6-27B-HauhauCS-Q4_K_P".into(),
            gguf: "Qwen3.6-27B-Uncensored-HauhauCS-Aggressive-Q4_K_P.gguf".into(),
            alias: "qwen3.6-27b-q4kp".into(),
            mmproj: "mmproj-Qwen3.6-27B-Uncensored-HauhauCS-Aggressive-f16.gguf".into(),
            name: "Qwen3.6-27B-HauhauCS-Q4_K_P".into(),
            ctx: 262_144,
            max_parallel: 2,
            kv_cache_type: "q8_0".into(),
            tensor_split: Some("24,24".into()),
            temperature: 1.0,
            thinking_policy: ThinkingPolicy::Disable,
            backend: BackendProfile::LlamaCpp,
        },
        QuantSpec {
            label: "Qwen3.6-27B-HauhauCS-Q5_K_P".into(),
            gguf: "Qwen3.6-27B-Uncensored-HauhauCS-Aggressive-Q5_K_P.gguf".into(),
            alias: "qwen3.6-27b-q5kp".into(),
            mmproj: "mmproj-Qwen3.6-27B-Uncensored-HauhauCS-Aggressive-f16.gguf".into(),
            name: "Qwen3.6-27B-HauhauCS-Q5_K_P".into(),
            ctx: 262_144,
            max_parallel: 2,
            kv_cache_type: "q8_0".into(),
            tensor_split: Some("24,24".into()),
            temperature: 1.0,
            thinking_policy: ThinkingPolicy::Disable,
            backend: BackendProfile::LlamaCpp,
        },
        QuantSpec {
            label: "Qwen3.6-27B-HauhauCS-Q6_K_P".into(),
            gguf: "Qwen3.6-27B-Uncensored-HauhauCS-Aggressive-Q6_K_P.gguf".into(),
            alias: "qwen3.6-27b-q6kp".into(),
            mmproj: "mmproj-Qwen3.6-27B-Uncensored-HauhauCS-Aggressive-f16.gguf".into(),
            name: "Qwen3.6-27B-HauhauCS-Q6_K_P".into(),
            ctx: 262_144,
            max_parallel: 2,
            kv_cache_type: "q8_0".into(),
            tensor_split: Some("24,24".into()),
            temperature: 1.0,
            thinking_policy: ThinkingPolicy::Disable,
            backend: BackendProfile::LlamaCpp,
        },
        QuantSpec {
            label: "Qwen3.6-27B-HauhauCS-Q8_K_P".into(),
            gguf: "Qwen3.6-27B-Uncensored-HauhauCS-Aggressive-Q8_K_P.gguf".into(),
            alias: "qwen3.6-27b-q8kp".into(),
            mmproj: "mmproj-Qwen3.6-27B-Uncensored-HauhauCS-Aggressive-f16.gguf".into(),
            name: "Qwen3.6-27B-HauhauCS-Q8_K_P".into(),
            ctx: 262_144,
            max_parallel: 1,
            kv_cache_type: "q8_0".into(),
            tensor_split: Some("24,24".into()),
            temperature: 1.0,
            thinking_policy: ThinkingPolicy::Disable,
            backend: BackendProfile::LlamaCpp,
        },
        // 35B-A3B baseline — note the relative paths point one dir up out of ~/models/qwen36-quants.
        QuantSpec {
            label: "Qwen3.6-35B-A3B-Q3_K_XL".into(),
            gguf: "../Qwen3.6-35B-A3B-UD-Q3_K_XL.gguf".into(),
            alias: "qwen3.6-35b-a3b".into(),
            mmproj: "../mmproj-F16.gguf".into(),
            name: "Qwen3.6-35B-A3B-Q3_K_XL".into(),
            ctx: 262_144,
            max_parallel: 1,
            kv_cache_type: "q8_0".into(),
            tensor_split: Some("24,24".into()),
            temperature: 1.0,
            thinking_policy: ThinkingPolicy::Disable,
            backend: BackendProfile::LlamaCpp,
        },
    ]
    .into_iter()
    .map(|q| (q.label.clone(), q))
    .collect()
});

/// Merge an external catalog (JSON mapping label -> QuantSpec) over `base`,
/// extending/overriding compiled entries. A parse error is logged and `base`
/// is returned unchanged (never panics).
fn apply_external_catalog(
    mut base: BTreeMap<String, QuantSpec>,
    json: &str,
) -> BTreeMap<String, QuantSpec> {
    match serde_json::from_str::<BTreeMap<String, QuantSpec>>(json) {
        Ok(extra) => {
            base.extend(extra);
            base
        }
        Err(e) => {
            eprintln!("warning: ignoring malformed SWEBENCH_QUANT_CATALOG: {e}");
            base
        }
    }
}

pub fn quant_catalog() -> BTreeMap<String, QuantSpec> {
    let base = CATALOG.clone();
    // Optional external catalog: SWEBENCH_QUANT_CATALOG points at a JSON file
    // mapping label -> QuantSpec, merged over the compiled defaults so
    // arbitrary local GGUFs can be used without editing this source.
    match std::env::var("SWEBENCH_QUANT_CATALOG") {
        Ok(path) if !path.is_empty() => match std::fs::read_to_string(&path) {
            Ok(text) => apply_external_catalog(base, &text),
            Err(e) => {
                eprintln!("warning: cannot read SWEBENCH_QUANT_CATALOG {path}: {e}");
                base
            }
        },
        _ => base,
    }
}

/// Default quant set used when `--quants` is not provided.
pub const DEFAULT_QUANTS: &[&str] = &[
    "Qwen3.6-35B-A3B-Q3_K_XL",
    "Qwen3.6-27B-HauhauCS-Q4_K_P",
    "Qwen3.6-27B-HauhauCS-IQ4_XS",
    "Qwen3.6-27B-HauhauCS-Q2_K_P",
];

#[cfg(test)]
#[path = "../../../tests/unit/bench_harness/swebench_pro/catalog/catalog_test.rs"]
mod tests;