From 5ff173fcc7ffa39f0c45b84fbdacefe55432320b Mon Sep 17 00:00:00 2001 From: SuYao Date: Sun, 28 Dec 2025 17:04:45 +0800 Subject: [PATCH] fix(ollama): improve reasoningEffort handling in providerOptions (#12089) * fix(ollama): improve reasoningEffort handling in providerOptions * fix(ollama): update reasoning effort handling and add support for gpt-oss models * fix(ollama): update think option to support 'low', 'medium', and 'high' values * fix(ollama): update comment to clarify accepted reasoning effort values for gpt-oss models --- ...-ai-provider-v2-npm-1.5.5-8bef249af9.patch | 12 +++++----- src/renderer/src/aiCore/utils/options.ts | 6 +++-- .../config/models/__tests__/reasoning.test.ts | 22 ++++++++++++++++++- src/renderer/src/config/models/reasoning.ts | 5 +++++ src/renderer/src/types/index.ts | 1 + yarn.lock | 4 ++-- 6 files changed, 39 insertions(+), 11 deletions(-) diff --git a/.yarn/patches/ollama-ai-provider-v2-npm-1.5.5-8bef249af9.patch b/.yarn/patches/ollama-ai-provider-v2-npm-1.5.5-8bef249af9.patch index ea14381539..c306bef6e5 100644 --- a/.yarn/patches/ollama-ai-provider-v2-npm-1.5.5-8bef249af9.patch +++ b/.yarn/patches/ollama-ai-provider-v2-npm-1.5.5-8bef249af9.patch @@ -7,7 +7,7 @@ index 8dd9b498050dbecd8dd6b901acf1aa8ca38a49af..ed644349c9d38fe2a66b2fb44214f7c1 type OllamaChatModelId = "athene-v2" | "athene-v2:72b" | "aya-expanse" | "aya-expanse:8b" | "aya-expanse:32b" | "codegemma" | "codegemma:2b" | "codegemma:7b" | "codellama" | "codellama:7b" | "codellama:13b" | "codellama:34b" | "codellama:70b" | "codellama:code" | "codellama:python" | "command-r" | "command-r:35b" | "command-r-plus" | "command-r-plus:104b" | "command-r7b" | "command-r7b:7b" | "deepseek-r1" | "deepseek-r1:1.5b" | "deepseek-r1:7b" | "deepseek-r1:8b" | "deepseek-r1:14b" | "deepseek-r1:32b" | "deepseek-r1:70b" | "deepseek-r1:671b" | "deepseek-coder-v2" | "deepseek-coder-v2:16b" | "deepseek-coder-v2:236b" | "deepseek-v3" | "deepseek-v3:671b" | "devstral" | "devstral:24b" | "dolphin3" | "dolphin3:8b" | "exaone3.5" | "exaone3.5:2.4b" | "exaone3.5:7.8b" | "exaone3.5:32b" | "falcon2" | "falcon2:11b" | "falcon3" | "falcon3:1b" | "falcon3:3b" | "falcon3:7b" | "falcon3:10b" | "firefunction-v2" | "firefunction-v2:70b" | "gemma" | "gemma:2b" | "gemma:7b" | "gemma2" | "gemma2:2b" | "gemma2:9b" | "gemma2:27b" | "gemma3" | "gemma3:1b" | "gemma3:4b" | "gemma3:12b" | "gemma3:27b" | "granite3-dense" | "granite3-dense:2b" | "granite3-dense:8b" | "granite3-guardian" | "granite3-guardian:2b" | "granite3-guardian:8b" | "granite3-moe" | "granite3-moe:1b" | "granite3-moe:3b" | "granite3.1-dense" | "granite3.1-dense:2b" | "granite3.1-dense:8b" | "granite3.1-moe" | "granite3.1-moe:1b" | "granite3.1-moe:3b" | "llama2" | "llama2:7b" | "llama2:13b" | "llama2:70b" | "llama3" | "llama3:8b" | "llama3:70b" | "llama3-chatqa" | "llama3-chatqa:8b" | "llama3-chatqa:70b" | "llama3-gradient" | "llama3-gradient:8b" | "llama3-gradient:70b" | "llama3.1" | "llama3.1:8b" | "llama3.1:70b" | "llama3.1:405b" | "llama3.2" | "llama3.2:1b" | "llama3.2:3b" | "llama3.2-vision" | "llama3.2-vision:11b" | "llama3.2-vision:90b" | "llama3.3" | "llama3.3:70b" | "llama4" | "llama4:16x17b" | "llama4:128x17b" | "llama-guard3" | "llama-guard3:1b" | "llama-guard3:8b" | "llava" | "llava:7b" | "llava:13b" | "llava:34b" | "llava-llama3" | "llava-llama3:8b" | "llava-phi3" | "llava-phi3:3.8b" | "marco-o1" | "marco-o1:7b" | "mistral" | "mistral:7b" | "mistral-large" | "mistral-large:123b" | "mistral-nemo" | "mistral-nemo:12b" | "mistral-small" | "mistral-small:22b" | "mixtral" | "mixtral:8x7b" | "mixtral:8x22b" | "moondream" | "moondream:1.8b" | "openhermes" | "openhermes:v2.5" | "nemotron" | "nemotron:70b" | "nemotron-mini" | "nemotron-mini:4b" | "olmo" | "olmo:7b" | "olmo:13b" | "opencoder" | "opencoder:1.5b" | "opencoder:8b" | "phi3" | "phi3:3.8b" | "phi3:14b" | "phi3.5" | "phi3.5:3.8b" | "phi4" | "phi4:14b" | "qwen" | "qwen:7b" | "qwen:14b" | "qwen:32b" | "qwen:72b" | "qwen:110b" | "qwen2" | "qwen2:0.5b" | "qwen2:1.5b" | "qwen2:7b" | "qwen2:72b" | "qwen2.5" | "qwen2.5:0.5b" | "qwen2.5:1.5b" | "qwen2.5:3b" | "qwen2.5:7b" | "qwen2.5:14b" | "qwen2.5:32b" | "qwen2.5:72b" | "qwen2.5-coder" | "qwen2.5-coder:0.5b" | "qwen2.5-coder:1.5b" | "qwen2.5-coder:3b" | "qwen2.5-coder:7b" | "qwen2.5-coder:14b" | "qwen2.5-coder:32b" | "qwen3" | "qwen3:0.6b" | "qwen3:1.7b" | "qwen3:4b" | "qwen3:8b" | "qwen3:14b" | "qwen3:30b" | "qwen3:32b" | "qwen3:235b" | "qwq" | "qwq:32b" | "sailor2" | "sailor2:1b" | "sailor2:8b" | "sailor2:20b" | "shieldgemma" | "shieldgemma:2b" | "shieldgemma:9b" | "shieldgemma:27b" | "smallthinker" | "smallthinker:3b" | "smollm" | "smollm:135m" | "smollm:360m" | "smollm:1.7b" | "tinyllama" | "tinyllama:1.1b" | "tulu3" | "tulu3:8b" | "tulu3:70b" | (string & {}); declare const ollamaProviderOptions: z.ZodObject<{ - think: z.ZodOptional; -+ think: z.ZodOptional]>>; ++ think: z.ZodOptional, z.ZodLiteral<"medium">, z.ZodLiteral<"high">]>>; options: z.ZodOptional; repeat_last_n: z.ZodOptional; @@ -29,7 +29,7 @@ index 8dd9b498050dbecd8dd6b901acf1aa8ca38a49af..ed644349c9d38fe2a66b2fb44214f7c1 declare const ollamaCompletionProviderOptions: z.ZodObject<{ - think: z.ZodOptional; -+ think: z.ZodOptional]>>; ++ think: z.ZodOptional, z.ZodLiteral<"medium">, z.ZodLiteral<"high">]>>; user: z.ZodOptional; suffix: z.ZodOptional; echo: z.ZodOptional; @@ -42,7 +42,7 @@ index 35b5142ce8476ce2549ed7c2ec48e7d8c46c90d9..2ef64dc9a4c2be043e6af608241a6a83 // src/completion/ollama-completion-language-model.ts var ollamaCompletionProviderOptions = import_v42.z.object({ - think: import_v42.z.boolean().optional(), -+ think: import_v42.z.union([import_v42.z.boolean(), import_v42.z.enum(['low', 'medium', 'high'])]).optional(), ++ think: import_v42.z.union([import_v42.z.boolean(), import_v42.z.literal('low'), import_v42.z.literal('medium'), import_v42.z.literal('high')]).optional(), user: import_v42.z.string().optional(), suffix: import_v42.z.string().optional(), echo: import_v42.z.boolean().optional() @@ -64,7 +64,7 @@ index 35b5142ce8476ce2549ed7c2ec48e7d8c46c90d9..2ef64dc9a4c2be043e6af608241a6a83 * Only supported by certain models like DeepSeek R1 and Qwen 3. */ - think: import_v44.z.boolean().optional(), -+ think: import_v44.z.union([import_v44.z.boolean(), import_v44.z.enum(['low', 'medium', 'high'])]).optional(), ++ think: import_v44.z.union([import_v44.z.boolean(), import_v44.z.literal('low'), import_v44.z.literal('medium'), import_v44.z.literal('high')]).optional(), options: import_v44.z.object({ num_ctx: import_v44.z.number().optional(), repeat_last_n: import_v44.z.number().optional(), @@ -97,7 +97,7 @@ index e2a634a78d80ac9542f2cc4f96cf2291094b10cf..67b23efce3c1cf4f026693d3ff924698 // src/completion/ollama-completion-language-model.ts var ollamaCompletionProviderOptions = z2.object({ - think: z2.boolean().optional(), -+ think: z2.union([z2.boolean(), z2.enum(['low', 'medium', 'high'])]).optional(), ++ think: z2.union([z2.boolean(), z2.literal('low'), z2.literal('medium'), z2.literal('high')]).optional(), user: z2.string().optional(), suffix: z2.string().optional(), echo: z2.boolean().optional() @@ -119,7 +119,7 @@ index e2a634a78d80ac9542f2cc4f96cf2291094b10cf..67b23efce3c1cf4f026693d3ff924698 * Only supported by certain models like DeepSeek R1 and Qwen 3. */ - think: z4.boolean().optional(), -+ think: z4.union([z4.boolean(), z4.enum(['low', 'medium', 'high'])]).optional(), ++ think: z4.union([z4.boolean(), z4.literal('low'), z4.literal('medium'), z4.literal('high')]).optional(), options: z4.object({ num_ctx: z4.number().optional(), repeat_last_n: z4.number().optional(), diff --git a/src/renderer/src/aiCore/utils/options.ts b/src/renderer/src/aiCore/utils/options.ts index 36778b7570..8dc7a10af9 100644 --- a/src/renderer/src/aiCore/utils/options.ts +++ b/src/renderer/src/aiCore/utils/options.ts @@ -580,8 +580,10 @@ function buildOllamaProviderOptions( const reasoningEffort = assistant.settings?.reasoning_effort if (enableReasoning) { if (isOpenAIOpenWeightModel(model)) { - // @ts-ignore upstream type error - providerOptions.think = reasoningEffort as any + // For gpt-oss models, Ollama accepts: 'low' | 'medium' | 'high' + if (reasoningEffort === 'low' || reasoningEffort === 'medium' || reasoningEffort === 'high') { + providerOptions.think = reasoningEffort + } } else { providerOptions.think = !['none', undefined].includes(reasoningEffort) } diff --git a/src/renderer/src/config/models/__tests__/reasoning.test.ts b/src/renderer/src/config/models/__tests__/reasoning.test.ts index 6b00a8912b..56f9cd0b60 100644 --- a/src/renderer/src/config/models/__tests__/reasoning.test.ts +++ b/src/renderer/src/config/models/__tests__/reasoning.test.ts @@ -680,7 +680,12 @@ describe('getThinkModelType - Comprehensive Coverage', () => { expect(getThinkModelType(createModel({ id: 'o3' }))).toBe('o') expect(getThinkModelType(createModel({ id: 'o3-mini' }))).toBe('o') expect(getThinkModelType(createModel({ id: 'o4' }))).toBe('o') - expect(getThinkModelType(createModel({ id: 'gpt-oss-reasoning' }))).toBe('o') + }) + + it('should return gpt_oss for gpt-oss models', () => { + expect(getThinkModelType(createModel({ id: 'gpt-oss' }))).toBe('gpt_oss') + expect(getThinkModelType(createModel({ id: 'gpt-oss:20b' }))).toBe('gpt_oss') + expect(getThinkModelType(createModel({ id: 'gpt-oss-reasoning' }))).toBe('gpt_oss') }) }) @@ -1763,6 +1768,21 @@ describe('getModelSupportedReasoningEffortOptions', () => { 'medium', 'high' ]) + }) + + it('should return correct options for gpt-oss models', () => { + expect(getModelSupportedReasoningEffortOptions(createModel({ id: 'gpt-oss' }))).toEqual([ + 'default', + 'low', + 'medium', + 'high' + ]) + expect(getModelSupportedReasoningEffortOptions(createModel({ id: 'gpt-oss:20b' }))).toEqual([ + 'default', + 'low', + 'medium', + 'high' + ]) expect(getModelSupportedReasoningEffortOptions(createModel({ id: 'gpt-oss-reasoning' }))).toEqual([ 'default', 'low', diff --git a/src/renderer/src/config/models/reasoning.ts b/src/renderer/src/config/models/reasoning.ts index 5d48e9a122..b2b6119b76 100644 --- a/src/renderer/src/config/models/reasoning.ts +++ b/src/renderer/src/config/models/reasoning.ts @@ -17,6 +17,7 @@ import { isGPT52ProModel, isGPT52SeriesModel, isOpenAIDeepResearchModel, + isOpenAIOpenWeightModel, isOpenAIReasoningModel, isSupportedReasoningEffortOpenAIModel } from './openai' @@ -41,6 +42,7 @@ export const MODEL_SUPPORTED_REASONING_EFFORT = { gpt5_2: ['none', 'low', 'medium', 'high', 'xhigh'] as const, gpt5pro: ['high'] as const, gpt52pro: ['medium', 'high', 'xhigh'] as const, + gpt_oss: ['low', 'medium', 'high'] as const, grok: ['low', 'high'] as const, grok4_fast: ['auto'] as const, gemini2_flash: ['low', 'medium', 'high', 'auto'] as const, @@ -72,6 +74,7 @@ export const MODEL_SUPPORTED_OPTIONS: ThinkingOptionConfig = { gpt5_2: ['default', ...MODEL_SUPPORTED_REASONING_EFFORT.gpt5_2] as const, gpt5_1_codex_max: ['default', ...MODEL_SUPPORTED_REASONING_EFFORT.gpt5_1_codex_max] as const, gpt52pro: ['default', ...MODEL_SUPPORTED_REASONING_EFFORT.gpt52pro] as const, + gpt_oss: ['default', ...MODEL_SUPPORTED_REASONING_EFFORT.gpt_oss] as const, grok: ['default', ...MODEL_SUPPORTED_REASONING_EFFORT.grok] as const, grok4_fast: ['default', 'none', ...MODEL_SUPPORTED_REASONING_EFFORT.grok4_fast] as const, gemini2_flash: ['default', 'none', ...MODEL_SUPPORTED_REASONING_EFFORT.gemini2_flash] as const, @@ -127,6 +130,8 @@ const _getThinkModelType = (model: Model): ThinkingModelType => { thinkingModelType = 'gpt5pro' } } + } else if (isOpenAIOpenWeightModel(model)) { + thinkingModelType = 'gpt_oss' } else if (isSupportedReasoningEffortOpenAIModel(model)) { thinkingModelType = 'o' } else if (isGrok4FastReasoningModel(model)) { diff --git a/src/renderer/src/types/index.ts b/src/renderer/src/types/index.ts index a75fc1ed3e..b87271930b 100644 --- a/src/renderer/src/types/index.ts +++ b/src/renderer/src/types/index.ts @@ -92,6 +92,7 @@ const ThinkModelTypes = [ 'gpt5_2', 'gpt5pro', 'gpt52pro', + 'gpt_oss', 'grok', 'grok4_fast', 'gemini2_flash', diff --git a/yarn.lock b/yarn.lock index b6b87c568a..3fe6d2c151 100644 --- a/yarn.lock +++ b/yarn.lock @@ -20209,13 +20209,13 @@ __metadata: "ollama-ai-provider-v2@patch:ollama-ai-provider-v2@npm%3A1.5.5#~/.yarn/patches/ollama-ai-provider-v2-npm-1.5.5-8bef249af9.patch": version: 1.5.5 - resolution: "ollama-ai-provider-v2@patch:ollama-ai-provider-v2@npm%3A1.5.5#~/.yarn/patches/ollama-ai-provider-v2-npm-1.5.5-8bef249af9.patch::version=1.5.5&hash=16c016" + resolution: "ollama-ai-provider-v2@patch:ollama-ai-provider-v2@npm%3A1.5.5#~/.yarn/patches/ollama-ai-provider-v2-npm-1.5.5-8bef249af9.patch::version=1.5.5&hash=0aef28" dependencies: "@ai-sdk/provider": "npm:^2.0.0" "@ai-sdk/provider-utils": "npm:^3.0.17" peerDependencies: zod: ^4.0.16 - checksum: 10c0/aa6bd3415d08f7bbd1a3051f45b1cd3a8fa8bb01413e98de45e8888f64e6b12bca6e340453a3e82e4193ca5354397f524c6c0f7b3e9996d70f53c81374c69180 + checksum: 10c0/32ca1f543ee791ac96061a5f6d8899c00644eeb774b3b951ca1e3e3810b60753acacf8229b2c1ba099b25a01732c54e51e0df44d11f4d90ae201f483d41aa149 languageName: node linkType: hard