release/v1.6.1
claude/github-issue-40-0au5ua
claude/linked-issue-acp-tui-u4vvpa
claude/minor-release-cut-vkugow
claude/sigit-acp-local-chat-cx6380
claude/sigit-cloud-agent-expansion-reox0a
claude/tool-permission-system
claude/zen-feynman-0u78dk
development
feature/acp-cancel-request
feature/acp-session-list-cursor
feature/acp-v1-conformance
feature/agent-tools-multiedit-glob-todos-remember
feature/cargo-workspace
feature/hooks
feature/load-local-model-explicitly
feature/sigit-agent-profile
feature/sigit-code-cloud
feature/xcode-mcp-server
fix/acp-replay-loaded-session-history
fix/acp-replay-reasoning-and-attachments
fix/dsml-doubled-bar-markers
main
release/v1.5.10
release/v1.5.6
release/v1.6.1
| 1 | //! Model picker types and item construction, shared across platforms. |
| 2 | //! The unix-only TUI in `chat.rs` pulls from here. |
| 3 | |
| 4 | use onde::inference::GgufModelConfig; |
| 5 | |
| 6 | use crate::setup::DiscoveredModel; |
| 7 | |
| 8 | pub(crate) use crate::setup::ModelCacheHealth; |
| 9 | |
| 10 | // ── Types ───────────────────────────────────────────────────────────────────── |
| 11 | |
| 12 | #[derive(Clone, Copy, PartialEq, Eq, PartialOrd, Ord)] |
| 13 | pub(crate) enum ModelSource { |
| 14 | Onde, |
| 15 | HuggingFace, |
| 16 | /// not downloaded yet — selecting it triggers a download into the app-group cache. |
| 17 | Available, |
| 18 | Fallback, |
| 19 | /// a siGit Code Cloud tier (runs over the network, not on-device). |
| 20 | Cloud, |
| 21 | } |
| 22 | |
| 23 | /// The broad nature of where inference runs. The picker groups by this and the |
| 24 | /// `local_inference` setting decides which group is the active (highlighted) one. |
| 25 | #[derive(Clone, Copy, PartialEq, Eq)] |
| 26 | pub(crate) enum InferenceKind { |
| 27 | Local, |
| 28 | Cloud, |
| 29 | } |
| 30 | |
| 31 | impl ModelSource { |
| 32 | /// On-device sources are `Local`; the cloud tiers are `Cloud`. |
| 33 | pub(crate) fn kind(self) -> InferenceKind { |
| 34 | match self { |
| 35 | ModelSource::Cloud => InferenceKind::Cloud, |
| 36 | _ => InferenceKind::Local, |
| 37 | } |
| 38 | } |
| 39 | } |
| 40 | |
| 41 | /// The active inference mode, from the persisted `local_inference` setting. |
| 42 | pub(crate) fn active_inference_kind() -> InferenceKind { |
| 43 | if crate::settings::local_inference_enabled() { |
| 44 | InferenceKind::Local |
| 45 | } else { |
| 46 | InferenceKind::Cloud |
| 47 | } |
| 48 | } |
| 49 | |
| 50 | #[derive(Clone)] |
| 51 | pub(crate) struct ModelPickerItem { |
| 52 | pub(crate) display_name: String, |
| 53 | pub(crate) description: String, |
| 54 | pub(crate) tool_calling: bool, |
| 55 | pub(crate) max_tokens: u64, |
| 56 | pub(crate) context_window_tokens: u64, |
| 57 | pub(crate) config: GgufModelConfig, |
| 58 | pub(crate) source_label: String, |
| 59 | |
| 60 | pub(crate) source: ModelSource, |
| 61 | pub(crate) cache_health: ModelCacheHealth, |
| 62 | /// `Some(tier)` for a siGit Code Cloud entry; `None` for an on-device model. |
| 63 | pub(crate) cloud_tier: Option<String>, |
| 64 | } |
| 65 | |
| 66 | // ── Model ID → GgufModelConfig mapping ──────────────────────────────────────── |
| 67 | |
| 68 | /// map a HF model ID to its config constructor, or `None` if we don't support it. |
| 69 | pub(crate) fn model_id_to_config(model_id: &str) -> Option<GgufModelConfig> { |
| 70 | Some(match model_id { |
| 71 | "bartowski/Qwen_Qwen3-4B-GGUF" => GgufModelConfig::qwen3_4b(), |
| 72 | "bartowski/Qwen_Qwen3-8B-GGUF" => GgufModelConfig::qwen3_8b(), |
| 73 | "bartowski/Qwen_Qwen3-14B-GGUF" => GgufModelConfig::qwen3_14b(), |
| 74 | "bartowski/Qwen_Qwen3-1.7B-GGUF" => GgufModelConfig::qwen3_1_7b(), |
| 75 | "bartowski/Qwen2.5-3B-Instruct-GGUF" => GgufModelConfig::qwen25_3b(), |
| 76 | "bartowski/Qwen2.5-1.5B-Instruct-GGUF" => GgufModelConfig::qwen25_1_5b(), |
| 77 | "bartowski/Qwen2.5-Coder-3B-Instruct-GGUF" => GgufModelConfig::qwen25_coder_3b(), |
| 78 | "bartowski/Qwen2.5-Coder-1.5B-Instruct-GGUF" => GgufModelConfig::qwen25_coder_1_5b(), |
| 79 | "bartowski/Qwen2.5-Coder-7B-Instruct-GGUF" => GgufModelConfig::qwen25_coder_7b(), |
| 80 | "TheBloke/deepseek-coder-6.7B-instruct-GGUF" => GgufModelConfig::deepseek_coder_6_7b(), |
| 81 | _ => return None, |
| 82 | }) |
| 83 | } |
| 84 | |
| 85 | fn is_tool_calling(model_id: &str) -> bool { |
| 86 | matches!( |
| 87 | model_id, |
| 88 | "bartowski/Qwen_Qwen3-4B-GGUF" |
| 89 | | "bartowski/Qwen_Qwen3-8B-GGUF" |
| 90 | | "bartowski/Qwen_Qwen3-14B-GGUF" |
| 91 | | "bartowski/Qwen_Qwen3-1.7B-GGUF" |
| 92 | | "bartowski/Qwen2.5-Coder-7B-Instruct-GGUF" |
| 93 | ) |
| 94 | } |
| 95 | |
| 96 | /// tool-calling models get more tokens because `<think>` blocks eat into the budget. |
| 97 | fn max_tokens_for(model_id: &str) -> u64 { |
| 98 | if is_tool_calling(model_id) { 4096 } else { 512 } |
| 99 | } |
| 100 | |
| 101 | /// Usable context window shown in editor model pickers. |
| 102 | /// |
| 103 | /// This is intentionally separate from `max_tokens`, which is the generation |
| 104 | /// cap. Qwen GGUFs are used at their default 32K window; Qwen2.5 can be stretched |
| 105 | /// further with YaRN in some deployments, but siGit Code does not configure that |
| 106 | /// here. |
| 107 | fn context_window_tokens_for(model_id: &str) -> u64 { |
| 108 | if model_id == "TheBloke/deepseek-coder-6.7B-instruct-GGUF" { |
| 109 | 16_000 |
| 110 | } else { |
| 111 | 32_768 |
| 112 | } |
| 113 | } |
| 114 | |
| 115 | /// Context window advertised for siGit Code Cloud tiers. |
| 116 | /// |
| 117 | /// Deliberately a value of its own rather than the compaction budget: the |
| 118 | /// budget is when siGit Code summarizes history, not how much the model can |
| 119 | /// hold. |
| 120 | /// Deriving the label from it made a user-facing number move whenever that |
| 121 | /// tuning constant changed, and understated the window by an order of |
| 122 | /// magnitude. |
| 123 | pub(crate) const CLOUD_CONTEXT_WINDOW_TOKENS: u64 = 200_000; |
| 124 | |
| 125 | /// Render a context window as a short badge, e.g. `24K ctx`. |
| 126 | /// |
| 127 | /// Shared by the TUI `/models` picker and the ACP session-config labels so a |
| 128 | /// model reads the same in both editors. |
| 129 | pub(crate) fn format_context_window(tokens: u64) -> String { |
| 130 | format!("{} ctx", format_context_window_short(tokens)) |
| 131 | } |
| 132 | |
| 133 | /// The same magnitude without the `ctx` suffix, for places that supply their |
| 134 | /// own wording (e.g. "200K context"). |
| 135 | pub(crate) fn format_context_window_short(tokens: u64) -> String { |
| 136 | if tokens >= 1_000 { |
| 137 | format!("{}K", tokens / 1_000) |
| 138 | } else { |
| 139 | tokens.to_string() |
| 140 | } |
| 141 | } |
| 142 | |
| 143 | // ── Builder ─────────────────────────────────────────────────────────────────── |
| 144 | |
| 145 | /// collect every model the picker should show: local cache, remote available, fallback. |
| 146 | /// sorted by source (Onde > HF > Available > Fallback), then alphabetically. |
| 147 | pub(crate) fn build_model_picker_items() -> Vec<ModelPickerItem> { |
| 148 | let mut items = Vec::new(); |
| 149 | |
| 150 | // ── 1. Locally discovered models ───────────────────────────────────── |
| 151 | for discovered in crate::setup::discover_local_models() { |
| 152 | if let Some(item) = discovered_model_to_picker_item(discovered) { |
| 153 | items.push(item); |
| 154 | } |
| 155 | } |
| 156 | |
| 157 | // ── 2. Supported models not yet downloaded ─────────────────────────── |
| 158 | for info in onde::inference::models::SUPPORTED_MODEL_INFO { |
| 159 | let already_present = items.iter().any(|item| item.config.model_id == info.id); |
| 160 | if already_present { |
| 161 | continue; |
| 162 | } |
| 163 | |
| 164 | let config = match model_id_to_config(info.id) { |
| 165 | Some(config) => config, |
| 166 | None => continue, |
| 167 | }; |
| 168 | |
| 169 | let tool_calling = is_tool_calling(info.id); |
| 170 | let max_tokens = max_tokens_for(info.id); |
| 171 | let context_window_tokens = context_window_tokens_for(info.id); |
| 172 | |
| 173 | items.push(ModelPickerItem { |
| 174 | display_name: config.display_name.clone(), |
| 175 | description: config.approx_memory.clone(), |
| 176 | tool_calling, |
| 177 | max_tokens, |
| 178 | context_window_tokens, |
| 179 | config, |
| 180 | source_label: "Onde".to_string(), |
| 181 | |
| 182 | source: ModelSource::Available, |
| 183 | cache_health: ModelCacheHealth::NotDownloaded, |
| 184 | cloud_tier: None, |
| 185 | }); |
| 186 | } |
| 187 | |
| 188 | // ── 3. Fallback (on-device default when nothing else is present) ───── |
| 189 | if items.is_empty() { |
| 190 | let config = GgufModelConfig::platform_default(); |
| 191 | let tool_calling = is_tool_calling(&config.model_id); |
| 192 | let max_tokens = max_tokens_for(&config.model_id); |
| 193 | let context_window_tokens = context_window_tokens_for(&config.model_id); |
| 194 | |
| 195 | items.push(ModelPickerItem { |
| 196 | display_name: config.display_name.clone(), |
| 197 | description: config.approx_memory.clone(), |
| 198 | tool_calling, |
| 199 | max_tokens, |
| 200 | context_window_tokens, |
| 201 | config, |
| 202 | source_label: "Platform default".to_string(), |
| 203 | |
| 204 | source: ModelSource::Fallback, |
| 205 | cache_health: ModelCacheHealth::Complete, |
| 206 | cloud_tier: None, |
| 207 | }); |
| 208 | } |
| 209 | |
| 210 | // ── 4. siGit Code Cloud tiers (always offered; sign-in gated at select) ─ |
| 211 | for tier in crate::provider::CLOUD_TIERS { |
| 212 | let label = crate::provider::cloud_tier_label(tier); |
| 213 | // Synthetic config: a `sigit-cloud:<tier>` id never collides with a real |
| 214 | // HuggingFace id (no `/`), so on-device matching code stays inert. |
| 215 | let config = GgufModelConfig { |
| 216 | model_id: format!("sigit-cloud:{tier}"), |
| 217 | files: Vec::new(), |
| 218 | tok_model_id: None, |
| 219 | display_name: label.clone(), |
| 220 | approx_memory: "Cloud".to_string(), |
| 221 | chat_template: None, |
| 222 | }; |
| 223 | items.push(ModelPickerItem { |
| 224 | display_name: label, |
| 225 | description: "siGit Code Cloud".to_string(), |
| 226 | tool_calling: true, |
| 227 | max_tokens: 4096, |
| 228 | context_window_tokens: CLOUD_CONTEXT_WINDOW_TOKENS, |
| 229 | config, |
| 230 | source_label: "siGit Code Cloud".to_string(), |
| 231 | source: ModelSource::Cloud, |
| 232 | cache_health: ModelCacheHealth::Complete, |
| 233 | cloud_tier: Some((*tier).to_string()), |
| 234 | }); |
| 235 | } |
| 236 | |
| 237 | // Group by inference nature, active mode first (so the highlighted group |
| 238 | // leads the picker), then by source, then alphabetically. |
| 239 | let active = active_inference_kind(); |
| 240 | let group_rank = |
| 241 | |item: &ModelPickerItem| -> u8 { if item.source.kind() == active { 0 } else { 1 } }; |
| 242 | items.sort_by(|left, right| { |
| 243 | group_rank(left) |
| 244 | .cmp(&group_rank(right)) |
| 245 | .then_with(|| left.source.cmp(&right.source)) |
| 246 | .then_with(|| left.display_name.cmp(&right.display_name)) |
| 247 | }); |
| 248 | |
| 249 | items |
| 250 | } |
| 251 | |
| 252 | /// Picker items restricted to on-device models (no cloud tiers). Used by the |
| 253 | /// model-loading and ACP session-config paths, which only handle local GGUF |
| 254 | /// models. The cloud tiers are an interactive TUI-picker feature. |
| 255 | pub(crate) fn local_picker_items() -> Vec<ModelPickerItem> { |
| 256 | build_model_picker_items() |
| 257 | .into_iter() |
| 258 | .filter(|item| item.cloud_tier.is_none()) |
| 259 | .collect() |
| 260 | } |
| 261 | |
| 262 | // ── Internal helpers ────────────────────────────────────────────────────────── |
| 263 | |
| 264 | fn discovered_model_to_picker_item(model: DiscoveredModel) -> Option<ModelPickerItem> { |
| 265 | let source_label = if model.from_app_group { |
| 266 | "Onde".to_string() |
| 267 | } else { |
| 268 | "HuggingFace".to_string() |
| 269 | }; |
| 270 | |
| 271 | let config = model_id_to_config(&model.model_id)?; |
| 272 | |
| 273 | let tool_calling = is_tool_calling(&model.model_id); |
| 274 | let max_tokens = max_tokens_for(&model.model_id); |
| 275 | let context_window_tokens = context_window_tokens_for(&model.model_id); |
| 276 | |
| 277 | Some(ModelPickerItem { |
| 278 | display_name: config.display_name.clone(), |
| 279 | description: config.approx_memory.clone(), |
| 280 | tool_calling, |
| 281 | max_tokens, |
| 282 | context_window_tokens, |
| 283 | config, |
| 284 | source_label, |
| 285 | |
| 286 | source: if model.from_app_group { |
| 287 | ModelSource::Onde |
| 288 | } else { |
| 289 | ModelSource::HuggingFace |
| 290 | }, |
| 291 | cache_health: model.cache_health, |
| 292 | cloud_tier: None, |
| 293 | }) |
| 294 | } |