| 1 | use std::collections::HashMap; |
| 2 | |
| 3 | use deepseek_config::ProviderKind; |
| 4 | use serde::{Deserialize, Serialize}; |
| 5 | |
| 6 | #[derive(Debug, Clone, Serialize, Deserialize)] |
| 7 | pub struct ModelInfo { |
| 8 | pub id: String, |
| 9 | pub provider: ProviderKind, |
| 10 | pub aliases: Vec<String>, |
| 11 | pub supports_tools: bool, |
| 12 | pub supports_reasoning: bool, |
| 13 | } |
| 14 | |
| 15 | #[derive(Debug, Clone, Serialize, Deserialize)] |
| 16 | pub struct ModelResolution { |
| 17 | pub requested: Option<String>, |
| 18 | pub resolved: ModelInfo, |
| 19 | pub used_fallback: bool, |
| 20 | pub fallback_chain: Vec<String>, |
| 21 | } |
| 22 | |
| 23 | #[derive(Debug, Clone)] |
| 24 | pub struct ModelRegistry { |
| 25 | models: Vec<ModelInfo>, |
| 26 | alias_map: HashMap<String, usize>, |
| 27 | } |
| 28 | |
| 29 | impl Default for ModelRegistry { |
| 30 | fn default() -> Self { |
| 31 | let models = vec![ |
| 32 | ModelInfo { |
| 33 | id: "deepseek-v4-pro".to_string(), |
| 34 | provider: ProviderKind::Deepseek, |
| 35 | aliases: vec![], |
| 36 | supports_tools: true, |
| 37 | supports_reasoning: true, |
| 38 | }, |
| 39 | ModelInfo { |
| 40 | id: "deepseek-v4-flash".to_string(), |
| 41 | provider: ProviderKind::Deepseek, |
| 42 | aliases: vec![ |
| 43 | "deepseek-chat".to_string(), |
| 44 | "deepseek-reasoner".to_string(), |
| 45 | "deepseek-r1".to_string(), |
| 46 | "deepseek-v3".to_string(), |
| 47 | "deepseek-v3.2".to_string(), |
| 48 | ], |
| 49 | supports_tools: true, |
| 50 | supports_reasoning: true, |
| 51 | }, |
| 52 | ModelInfo { |
| 53 | id: "deepseek-ai/deepseek-v4-pro".to_string(), |
| 54 | provider: ProviderKind::NvidiaNim, |
| 55 | aliases: vec![ |
| 56 | "deepseek-v4-pro".to_string(), |
| 57 | "nvidia-deepseek-v4-pro".to_string(), |
| 58 | "nim-deepseek-v4-pro".to_string(), |
| 59 | ], |
| 60 | supports_tools: true, |
| 61 | supports_reasoning: true, |
| 62 | }, |
| 63 | ModelInfo { |
| 64 | id: "deepseek-ai/deepseek-v4-flash".to_string(), |
| 65 | provider: ProviderKind::NvidiaNim, |
| 66 | aliases: vec![ |
| 67 | "deepseek-v4-flash".to_string(), |
| 68 | "deepseek-chat".to_string(), |
| 69 | "deepseek-reasoner".to_string(), |
| 70 | "nvidia-deepseek-v4-flash".to_string(), |
| 71 | "nim-deepseek-v4-flash".to_string(), |
| 72 | ], |
| 73 | supports_tools: true, |
| 74 | supports_reasoning: true, |
| 75 | }, |
| 76 | ModelInfo { |
| 77 | id: "gpt-4.1".to_string(), |
| 78 | provider: ProviderKind::Openai, |
| 79 | aliases: vec!["gpt4.1".to_string(), "gpt-4o".to_string()], |
| 80 | supports_tools: true, |
| 81 | supports_reasoning: true, |
| 82 | }, |
| 83 | ModelInfo { |
| 84 | id: "gpt-4.1-mini".to_string(), |
| 85 | provider: ProviderKind::Openai, |
| 86 | aliases: vec!["gpt-4o-mini".to_string()], |
| 87 | supports_tools: true, |
| 88 | supports_reasoning: false, |
| 89 | }, |
| 90 | ModelInfo { |
| 91 | id: "deepseek/deepseek-v4-pro".to_string(), |
| 92 | provider: ProviderKind::Openrouter, |
| 93 | aliases: vec![ |
| 94 | "deepseek-v4-pro".to_string(), |
| 95 | "openrouter-deepseek-v4-pro".to_string(), |
| 96 | ], |
| 97 | supports_tools: true, |
| 98 | supports_reasoning: true, |
| 99 | }, |
| 100 | ModelInfo { |
| 101 | id: "deepseek/deepseek-v4-flash".to_string(), |
| 102 | provider: ProviderKind::Openrouter, |
| 103 | aliases: vec![ |
| 104 | "deepseek-v4-flash".to_string(), |
| 105 | "deepseek-chat".to_string(), |
| 106 | "deepseek-reasoner".to_string(), |
| 107 | "openrouter-deepseek-v4-flash".to_string(), |
| 108 | ], |
| 109 | supports_tools: true, |
| 110 | supports_reasoning: true, |
| 111 | }, |
| 112 | ModelInfo { |
| 113 | id: "deepseek/deepseek-v4-pro".to_string(), |
| 114 | provider: ProviderKind::Novita, |
| 115 | aliases: vec![ |
| 116 | "deepseek-v4-pro".to_string(), |
| 117 | "novita-deepseek-v4-pro".to_string(), |
| 118 | ], |
| 119 | supports_tools: true, |
| 120 | supports_reasoning: true, |
| 121 | }, |
| 122 | ModelInfo { |
| 123 | id: "deepseek/deepseek-v4-flash".to_string(), |
| 124 | provider: ProviderKind::Novita, |
| 125 | aliases: vec![ |
| 126 | "deepseek-v4-flash".to_string(), |
| 127 | "deepseek-chat".to_string(), |
| 128 | "deepseek-reasoner".to_string(), |
| 129 | "novita-deepseek-v4-flash".to_string(), |
| 130 | ], |
| 131 | supports_tools: true, |
| 132 | supports_reasoning: true, |
| 133 | }, |
| 134 | ModelInfo { |
| 135 | id: "accounts/fireworks/models/deepseek-v4-pro".to_string(), |
| 136 | provider: ProviderKind::Fireworks, |
| 137 | aliases: vec![ |
| 138 | "deepseek-v4-pro".to_string(), |
| 139 | "fireworks-deepseek-v4-pro".to_string(), |
| 140 | ], |
| 141 | supports_tools: true, |
| 142 | supports_reasoning: true, |
| 143 | }, |
| 144 | ModelInfo { |
| 145 | id: "deepseek-ai/DeepSeek-V4-Pro".to_string(), |
| 146 | provider: ProviderKind::Sglang, |
| 147 | aliases: vec![ |
| 148 | "deepseek-v4-pro".to_string(), |
| 149 | "sglang-deepseek-v4-pro".to_string(), |
| 150 | ], |
| 151 | supports_tools: true, |
| 152 | supports_reasoning: true, |
| 153 | }, |
| 154 | ModelInfo { |
| 155 | id: "deepseek-ai/DeepSeek-V4-Flash".to_string(), |
| 156 | provider: ProviderKind::Sglang, |
| 157 | aliases: vec![ |
| 158 | "deepseek-v4-flash".to_string(), |
| 159 | "deepseek-chat".to_string(), |
| 160 | "deepseek-reasoner".to_string(), |
| 161 | "sglang-deepseek-v4-flash".to_string(), |
| 162 | ], |
| 163 | supports_tools: true, |
| 164 | supports_reasoning: true, |
| 165 | }, |
| 166 | ModelInfo { |
| 167 | id: "deepseek-ai/DeepSeek-V4-Pro".to_string(), |
| 168 | provider: ProviderKind::Vllm, |
| 169 | aliases: vec![ |
| 170 | "deepseek-v4-pro".to_string(), |
| 171 | "vllm-deepseek-v4-pro".to_string(), |
| 172 | ], |
| 173 | supports_tools: true, |
| 174 | supports_reasoning: true, |
| 175 | }, |
| 176 | ModelInfo { |
| 177 | id: "deepseek-ai/DeepSeek-V4-Flash".to_string(), |
| 178 | provider: ProviderKind::Vllm, |
| 179 | aliases: vec![ |
| 180 | "deepseek-v4-flash".to_string(), |
| 181 | "deepseek-chat".to_string(), |
| 182 | "deepseek-reasoner".to_string(), |
| 183 | "vllm-deepseek-v4-flash".to_string(), |
| 184 | ], |
| 185 | supports_tools: true, |
| 186 | supports_reasoning: true, |
| 187 | }, |
| 188 | ]; |
| 189 | Self::new(models) |
| 190 | } |
| 191 | } |
| 192 | |
| 193 | impl ModelRegistry { |
| 194 | #[must_use] |
| 195 | pub fn new(models: Vec<ModelInfo>) -> Self { |
| 196 | let mut alias_map = HashMap::new(); |
| 197 | for (idx, model) in models.iter().enumerate() { |
| 198 | alias_map.entry(normalize(&model.id)).or_insert(idx); |
| 199 | for alias in &model.aliases { |
| 200 | alias_map.entry(normalize(alias)).or_insert(idx); |
| 201 | } |
| 202 | } |
| 203 | Self { models, alias_map } |
| 204 | } |
| 205 | |
| 206 | #[must_use] |
| 207 | pub fn list(&self) -> Vec<ModelInfo> { |
| 208 | self.models.clone() |
| 209 | } |
| 210 | |
| 211 | #[must_use] |
| 212 | pub fn resolve( |
| 213 | &self, |
| 214 | requested: Option<&str>, |
| 215 | provider_hint: Option<ProviderKind>, |
| 216 | ) -> ModelResolution { |
| 217 | let mut fallback_chain = Vec::new(); |
| 218 | |
| 219 | if let Some(name) = requested { |
| 220 | fallback_chain.push(format!("requested:{name}")); |
| 221 | if let Some(provider) = provider_hint |
| 222 | && let Some(model) = self |
| 223 | .models |
| 224 | .iter() |
| 225 | .find(|m| m.provider == provider && model_matches(m, name)) |
| 226 | .cloned() |
| 227 | { |
| 228 | return ModelResolution { |
| 229 | requested: Some(name.to_string()), |
| 230 | resolved: preserve_requested_model_id_case(model, name), |
| 231 | used_fallback: false, |
| 232 | fallback_chain, |
| 233 | }; |
| 234 | } |
| 235 | if let Some(idx) = self.alias_map.get(&normalize(name)) { |
| 236 | return ModelResolution { |
| 237 | requested: Some(name.to_string()), |
| 238 | resolved: preserve_requested_model_id_case(self.models[*idx].clone(), name), |
| 239 | used_fallback: false, |
| 240 | fallback_chain, |
| 241 | }; |
| 242 | } |
| 243 | } |
| 244 | |
| 245 | let provider = provider_hint.unwrap_or(ProviderKind::Deepseek); |
| 246 | fallback_chain.push(format!("provider_default:{}", provider.as_str())); |
| 247 | if let Some(model) = self.models.iter().find(|m| m.provider == provider).cloned() { |
| 248 | return ModelResolution { |
| 249 | requested: requested.map(ToOwned::to_owned), |
| 250 | resolved: model, |
| 251 | used_fallback: true, |
| 252 | fallback_chain, |
| 253 | }; |
| 254 | } |
| 255 | |
| 256 | let final_fallback = self.models.first().cloned().unwrap_or(ModelInfo { |
| 257 | id: "deepseek-v4-pro".to_string(), |
| 258 | provider: ProviderKind::Deepseek, |
| 259 | aliases: Vec::new(), |
| 260 | supports_tools: true, |
| 261 | supports_reasoning: true, |
| 262 | }); |
| 263 | fallback_chain.push("global_default:deepseek-v4-pro".to_string()); |
| 264 | ModelResolution { |
| 265 | requested: requested.map(ToOwned::to_owned), |
| 266 | resolved: final_fallback, |
| 267 | used_fallback: true, |
| 268 | fallback_chain, |
| 269 | } |
| 270 | } |
| 271 | } |
| 272 | |
| 273 | fn normalize(value: &str) -> String { |
| 274 | value.trim().to_ascii_lowercase() |
| 275 | } |
| 276 | |
| 277 | fn model_matches(model: &ModelInfo, requested: &str) -> bool { |
| 278 | let requested = normalize(requested); |
| 279 | normalize(&model.id) == requested |
| 280 | || model |
| 281 | .aliases |
| 282 | .iter() |
| 283 | .any(|alias| normalize(alias) == requested) |
| 284 | } |
| 285 | |
| 286 | fn preserve_requested_model_id_case(mut model: ModelInfo, requested: &str) -> ModelInfo { |
| 287 | let requested = requested.trim(); |
| 288 | if model.id.eq_ignore_ascii_case(requested) { |
| 289 | model.id = requested.to_string(); |
| 290 | } |
| 291 | model |
| 292 | } |
| 293 | |
| 294 | #[cfg(test)] |
| 295 | mod tests { |
| 296 | use super::*; |
| 297 | |
| 298 | #[test] |
| 299 | fn deepseek_v4_pro_alias_stays_deepseek_by_default() { |
| 300 | let registry = ModelRegistry::default(); |
| 301 | let resolved = registry.resolve(Some("deepseek-v4-pro"), None); |
| 302 | |
| 303 | assert_eq!(resolved.resolved.provider, ProviderKind::Deepseek); |
| 304 | assert_eq!(resolved.resolved.id, "deepseek-v4-pro"); |
| 305 | } |
| 306 | |
| 307 | #[test] |
| 308 | fn deepseek_v4_pro_alias_resolves_to_nvidia_nim_when_provider_hinted() { |
| 309 | let registry = ModelRegistry::default(); |
| 310 | let resolved = registry.resolve(Some("deepseek-v4-pro"), Some(ProviderKind::NvidiaNim)); |
| 311 | |
| 312 | assert_eq!(resolved.resolved.provider, ProviderKind::NvidiaNim); |
| 313 | assert_eq!(resolved.resolved.id, "deepseek-ai/deepseek-v4-pro"); |
| 314 | } |
| 315 | |
| 316 | #[test] |
| 317 | fn nvidia_nim_default_uses_catalog_model_id() { |
| 318 | let registry = ModelRegistry::default(); |
| 319 | let resolved = registry.resolve(None, Some(ProviderKind::NvidiaNim)); |
| 320 | |
| 321 | assert_eq!(resolved.resolved.provider, ProviderKind::NvidiaNim); |
| 322 | assert_eq!(resolved.resolved.id, "deepseek-ai/deepseek-v4-pro"); |
| 323 | } |
| 324 | |
| 325 | #[test] |
| 326 | fn deepseek_v4_flash_alias_resolves_to_nvidia_nim_when_provider_hinted() { |
| 327 | let registry = ModelRegistry::default(); |
| 328 | let resolved = registry.resolve(Some("deepseek-v4-flash"), Some(ProviderKind::NvidiaNim)); |
| 329 | |
| 330 | assert_eq!(resolved.resolved.provider, ProviderKind::NvidiaNim); |
| 331 | assert_eq!(resolved.resolved.id, "deepseek-ai/deepseek-v4-flash"); |
| 332 | } |
| 333 | |
| 334 | #[test] |
| 335 | fn openrouter_default_uses_namespaced_model_id() { |
| 336 | let registry = ModelRegistry::default(); |
| 337 | let resolved = registry.resolve(None, Some(ProviderKind::Openrouter)); |
| 338 | |
| 339 | assert_eq!(resolved.resolved.provider, ProviderKind::Openrouter); |
| 340 | assert_eq!(resolved.resolved.id, "deepseek/deepseek-v4-pro"); |
| 341 | } |
| 342 | |
| 343 | #[test] |
| 344 | fn novita_default_uses_namespaced_model_id() { |
| 345 | let registry = ModelRegistry::default(); |
| 346 | let resolved = registry.resolve(None, Some(ProviderKind::Novita)); |
| 347 | |
| 348 | assert_eq!(resolved.resolved.provider, ProviderKind::Novita); |
| 349 | assert_eq!(resolved.resolved.id, "deepseek/deepseek-v4-pro"); |
| 350 | } |
| 351 | |
| 352 | #[test] |
| 353 | fn fireworks_default_uses_canonical_model_id() { |
| 354 | let registry = ModelRegistry::default(); |
| 355 | let resolved = registry.resolve(None, Some(ProviderKind::Fireworks)); |
| 356 | |
| 357 | assert_eq!(resolved.resolved.provider, ProviderKind::Fireworks); |
| 358 | assert_eq!( |
| 359 | resolved.resolved.id, |
| 360 | "accounts/fireworks/models/deepseek-v4-pro" |
| 361 | ); |
| 362 | } |
| 363 | |
| 364 | #[test] |
| 365 | fn sglang_default_uses_canonical_model_id() { |
| 366 | let registry = ModelRegistry::default(); |
| 367 | let resolved = registry.resolve(None, Some(ProviderKind::Sglang)); |
| 368 | |
| 369 | assert_eq!(resolved.resolved.provider, ProviderKind::Sglang); |
| 370 | assert_eq!(resolved.resolved.id, "deepseek-ai/DeepSeek-V4-Pro"); |
| 371 | } |
| 372 | |
| 373 | #[test] |
| 374 | fn deepseek_v4_flash_alias_resolves_to_openrouter_when_provider_hinted() { |
| 375 | let registry = ModelRegistry::default(); |
| 376 | let resolved = registry.resolve(Some("deepseek-v4-flash"), Some(ProviderKind::Openrouter)); |
| 377 | |
| 378 | assert_eq!(resolved.resolved.provider, ProviderKind::Openrouter); |
| 379 | assert_eq!(resolved.resolved.id, "deepseek/deepseek-v4-flash"); |
| 380 | } |
| 381 | |
| 382 | #[test] |
| 383 | fn deepseek_v4_flash_alias_resolves_to_novita_when_provider_hinted() { |
| 384 | let registry = ModelRegistry::default(); |
| 385 | let resolved = registry.resolve(Some("deepseek-v4-flash"), Some(ProviderKind::Novita)); |
| 386 | |
| 387 | assert_eq!(resolved.resolved.provider, ProviderKind::Novita); |
| 388 | assert_eq!(resolved.resolved.id, "deepseek/deepseek-v4-flash"); |
| 389 | } |
| 390 | |
| 391 | #[test] |
| 392 | fn deepseek_v4_flash_alias_resolves_to_sglang_when_provider_hinted() { |
| 393 | let registry = ModelRegistry::default(); |
| 394 | let resolved = registry.resolve(Some("deepseek-v4-flash"), Some(ProviderKind::Sglang)); |
| 395 | |
| 396 | assert_eq!(resolved.resolved.provider, ProviderKind::Sglang); |
| 397 | assert_eq!(resolved.resolved.id, "deepseek-ai/DeepSeek-V4-Flash"); |
| 398 | } |
| 399 | |
| 400 | #[test] |
| 401 | fn vllm_default_uses_canonical_model_id() { |
| 402 | let registry = ModelRegistry::default(); |
| 403 | let resolved = registry.resolve(None, Some(ProviderKind::Vllm)); |
| 404 | |
| 405 | assert_eq!(resolved.resolved.provider, ProviderKind::Vllm); |
| 406 | assert_eq!(resolved.resolved.id, "deepseek-ai/DeepSeek-V4-Pro"); |
| 407 | } |
| 408 | |
| 409 | #[test] |
| 410 | fn deepseek_v4_flash_alias_resolves_to_vllm_when_provider_hinted() { |
| 411 | let registry = ModelRegistry::default(); |
| 412 | let resolved = registry.resolve(Some("deepseek-v4-flash"), Some(ProviderKind::Vllm)); |
| 413 | |
| 414 | assert_eq!(resolved.resolved.provider, ProviderKind::Vllm); |
| 415 | assert_eq!(resolved.resolved.id, "deepseek-ai/DeepSeek-V4-Flash"); |
| 416 | } |
| 417 | |
| 418 | #[test] |
| 419 | fn preserves_requested_model_casing_for_third_party_providers() { |
| 420 | let registry = ModelRegistry::default(); |
| 421 | let resolved = registry.resolve(Some("DeepSeek-V4-Pro"), None); |
| 422 | |
| 423 | assert_eq!(resolved.resolved.provider, ProviderKind::Deepseek); |
| 424 | assert_eq!(resolved.resolved.id, "DeepSeek-V4-Pro"); |
| 425 | } |
| 426 | |
| 427 | #[test] |
| 428 | fn preserves_requested_model_casing_with_provider_hint() { |
| 429 | let registry = ModelRegistry::default(); |
| 430 | let resolved = registry.resolve(Some("DeepSeek-V4-Pro"), Some(ProviderKind::Deepseek)); |
| 431 | |
| 432 | assert_eq!(resolved.resolved.provider, ProviderKind::Deepseek); |
| 433 | assert_eq!(resolved.resolved.id, "DeepSeek-V4-Pro"); |
| 434 | } |
| 435 | |
| 436 | #[test] |
| 437 | fn preserves_requested_model_casing_without_surrounding_whitespace() { |
| 438 | let registry = ModelRegistry::default(); |
| 439 | let resolved = registry.resolve(Some(" DeepSeek-V4-Pro "), None); |
| 440 | |
| 441 | assert_eq!(resolved.resolved.provider, ProviderKind::Deepseek); |
| 442 | assert_eq!(resolved.resolved.id, "DeepSeek-V4-Pro"); |
| 443 | } |
| 444 | |
| 445 | #[test] |
| 446 | fn alias_match_does_not_override_requested_casing() { |
| 447 | let registry = ModelRegistry::default(); |
| 448 | let resolved = registry.resolve(Some("deepseek-reasoner"), None); |
| 449 | |
| 450 | assert_eq!(resolved.resolved.provider, ProviderKind::Deepseek); |
| 451 | assert_eq!(resolved.resolved.id, "deepseek-v4-flash"); |
| 452 | } |
| 453 | } |
| 454 |