fix context length for lmstudio and ollama (#2462)

packages/types/src/providers/index.ts

@@ -8,6 +8,7 @@ export * from "./groq.js"
 export * from "./lite-llm.js"
 export * from "./lm-studio.js"
 export * from "./mistral.js"
+export * from "./ollama.js"
 export * from "./openai.js"
 export * from "./openrouter.js"
 export * from "./requesty.js"

packages/types/src/providers/lm-studio.ts

@@ -1 +1,19 @@
+import type { ModelInfo } from "../model.js"
+
 export const LMSTUDIO_DEFAULT_TEMPERATURE = 0
+
+// LM Studio
+// https://lmstudio.ai/docs/cli/ls
+export const lMStudioDefaultModelId = "mistralai/devstral-small-2505"
+export const lMStudioDefaultModelInfo: ModelInfo = {
+	maxTokens: 8192,
+	contextWindow: 200_000,
+	supportsImages: true,
+	supportsComputerUse: true,
+	supportsPromptCache: true,
+	inputPrice: 0,
+	outputPrice: 0,
+	cacheWritesPrice: 0,
+	cacheReadsPrice: 0,
+	description: "LM Studio hosted models",
+}

packages/types/src/providers/ollama.ts

@@ -0,0 +1,17 @@
+import type { ModelInfo } from "../model.js"
+
+// Ollama
+// https://ollama.com/models
+export const ollamaDefaultModelId = "devstral:24b"
+export const ollamaDefaultModelInfo: ModelInfo = {
+	maxTokens: 4096,
+	contextWindow: 200_000,
+	supportsImages: true,
+	supportsComputerUse: true,
+	supportsPromptCache: true,
+	inputPrice: 0,
+	outputPrice: 0,
+	cacheWritesPrice: 0,
+	cacheReadsPrice: 0,
+	description: "Ollama hosted models",
+}

src/api/providers/fetchers/__tests__/fixtures/lmstudio-model-details.json

@@ -0,0 +1,14 @@
+{
+	"mistralai/devstral-small-2505": {
+		"type": "llm",
+		"modelKey": "mistralai/devstral-small-2505",
+		"format": "safetensors",
+		"displayName": "Devstral Small 2505",
+		"path": "mistralai/devstral-small-2505",
+		"sizeBytes": 13277565112,
+		"architecture": "mistral",
+		"vision": false,
+		"trainedForToolUse": false,
+		"maxContextLength": 131072
+	}
+}

src/api/providers/fetchers/__tests__/fixtures/ollama-model-details.json

@@ -0,0 +1,58 @@
+{
+	"qwen3-2to16:latest": {
+		"license": "                                 Apache License\\n                           Version 2.0, January 2004\\n...",
+		"modelfile": "model.modelfile,# To build a new Modelfile based on this, replace FROM with:...",
+		"parameters": "repeat_penalty                 1\\nstop                           \\\\nstop...",
+		"template": "{{- if .Messages }}\\n{{- if or .System .Tools }}<|im_start|>system...",
+		"details": {
+			"parent_model": "/Users/brad/.ollama/models/blobs/sha256-3291abe70f16ee9682de7bfae08db5373ea9d6497e614aaad63340ad421d6312",
+			"format": "gguf",
+			"family": "qwen3",
+			"families": ["qwen3"],
+			"parameter_size": "32.8B",
+			"quantization_level": "Q4_K_M"
+		},
+		"model_info": {
+			"general.architecture": "qwen3",
+			"general.basename": "Qwen3",
+			"general.file_type": 15,
+			"general.parameter_count": 32762123264,
+			"general.quantization_version": 2,
+			"general.size_label": "32B",
+			"general.type": "model",
+			"qwen3.attention.head_count": 64,
+			"qwen3.attention.head_count_kv": 8,
+			"qwen3.attention.key_length": 128,
+			"qwen3.attention.layer_norm_rms_epsilon": 0.000001,
+			"qwen3.attention.value_length": 128,
+			"qwen3.block_count": 64,
+			"qwen3.context_length": 40960,
+			"qwen3.embedding_length": 5120,
+			"qwen3.feed_forward_length": 25600,
+			"qwen3.rope.freq_base": 1000000,
+			"tokenizer.ggml.add_bos_token": false,
+			"tokenizer.ggml.bos_token_id": 151643,
+			"tokenizer.ggml.eos_token_id": 151645,
+			"tokenizer.ggml.merges": null,
+			"tokenizer.ggml.model": "gpt2",
+			"tokenizer.ggml.padding_token_id": 151643,
+			"tokenizer.ggml.pre": "qwen2",
+			"tokenizer.ggml.token_type": null,
+			"tokenizer.ggml.tokens": null
+		},
+		"tensors": [
+			{
+				"name": "output.weight",
+				"type": "Q6_K",
+				"shape": [5120, 151936]
+			},
+			{
+				"name": "output_norm.weight",
+				"type": "F32",
+				"shape": [5120]
+			}
+		],
+		"capabilities": ["completion", "tools"],
+		"modified_at": "2025-06-02T22:16:13.644123606-04:00"
+	}
+}

src/api/providers/fetchers/__tests__/lmstudio.test.ts

@@ -0,0 +1,197 @@
+import axios from "axios"
+import { vi, describe, it, expect, beforeEach } from "vitest"
+import { LMStudioClient, LLM, LLMInstanceInfo } from "@lmstudio/sdk" // LLMInfo is a type
+import { getLMStudioModels, parseLMStudioModel } from "../lmstudio"
+import { ModelInfo, lMStudioDefaultModelInfo } from "@roo-code/types" // ModelInfo is a type
+
+// Mock axios
+vi.mock("axios")
+const mockedAxios = axios as any
+
+// Mock @lmstudio/sdk
+const mockGetModelInfo = vi.fn()
+const mockListLoaded = vi.fn()
+vi.mock("@lmstudio/sdk", () => {
+	return {
+		LMStudioClient: vi.fn().mockImplementation(() => ({
+			llm: {
+				listLoaded: mockListLoaded,
+			},
+		})),
+	}
+})
+const MockedLMStudioClientConstructor = LMStudioClient as any
+
+describe("LMStudio Fetcher", () => {
+	beforeEach(() => {
+		vi.clearAllMocks()
+		MockedLMStudioClientConstructor.mockClear()
+		mockListLoaded.mockClear()
+		mockGetModelInfo.mockClear()
+	})
+
+	describe("parseLMStudioModel", () => {
+		it("should correctly parse raw LLMInfo to ModelInfo", () => {
+			const rawModel: LLMInstanceInfo = {
+				type: "llm",
+				modelKey: "mistralai/devstral-small-2505",
+				format: "safetensors",
+				displayName: "Devstral Small 2505",
+				path: "mistralai/devstral-small-2505",
+				sizeBytes: 13277565112,
+				architecture: "mistral",
+				identifier: "mistralai/devstral-small-2505",
+				instanceReference: "RAP5qbeHVjJgBiGFQ6STCuTJ",
+				vision: false,
+				trainedForToolUse: false,
+				maxContextLength: 131072,
+				contextLength: 7161,
+			}
+
+			const expectedModelInfo: ModelInfo = {
+				...lMStudioDefaultModelInfo,
+				description: `${rawModel.displayName} - ${rawModel.path}`,
+				contextWindow: rawModel.contextLength,
+				supportsPromptCache: true,
+				supportsImages: rawModel.vision,
+				supportsComputerUse: false,
+				maxTokens: rawModel.contextLength,
+				inputPrice: 0,
+				outputPrice: 0,
+				cacheWritesPrice: 0,
+				cacheReadsPrice: 0,
+			}
+
+			const result = parseLMStudioModel(rawModel)
+			expect(result).toEqual(expectedModelInfo)
+		})
+	})
+
+	describe("getLMStudioModels", () => {
+		const baseUrl = "http://localhost:1234"
+		const lmsUrl = "ws://localhost:1234"
+
+		const mockRawModel: LLMInstanceInfo = {
+			architecture: "test-arch",
+			identifier: "mistralai/devstral-small-2505",
+			instanceReference: "RAP5qbeHVjJgBiGFQ6STCuTJ",
+			modelKey: "test-model-key-1",
+			path: "/path/to/test-model-1",
+			type: "llm",
+			displayName: "Test Model One",
+			maxContextLength: 2048,
+			contextLength: 7161,
+			paramsString: "1B params, 2k context",
+			vision: true,
+			format: "gguf",
+			sizeBytes: 1000000000,
+			trainedForToolUse: false, // Added
+		}
+
+		it("should fetch and parse models successfully", async () => {
+			mockedAxios.get.mockResolvedValueOnce({ data: { status: "ok" } })
+			mockListLoaded.mockResolvedValueOnce([{ getModelInfo: mockGetModelInfo }])
+			mockGetModelInfo.mockResolvedValueOnce(mockRawModel)
+
+			const result = await getLMStudioModels(baseUrl)
+
+			expect(mockedAxios.get).toHaveBeenCalledTimes(1)
+			expect(mockedAxios.get).toHaveBeenCalledWith(`${baseUrl}/v1/models`)
+			expect(MockedLMStudioClientConstructor).toHaveBeenCalledTimes(1)
+			expect(MockedLMStudioClientConstructor).toHaveBeenCalledWith({ baseUrl: lmsUrl })
+			expect(mockListLoaded).toHaveBeenCalledTimes(1)
+
+			const expectedParsedModel = parseLMStudioModel(mockRawModel)
+			expect(result).toEqual({ [mockRawModel.modelKey]: expectedParsedModel })
+		})
+
+		it("should use default baseUrl if an empty string is provided", async () => {
+			const defaultBaseUrl = "http://localhost:1234"
+			const defaultLmsUrl = "ws://localhost:1234"
+			mockedAxios.get.mockResolvedValueOnce({ data: {} })
+			mockListLoaded.mockResolvedValueOnce([])
+
+			await getLMStudioModels("")
+
+			expect(mockedAxios.get).toHaveBeenCalledWith(`${defaultBaseUrl}/v1/models`)
+			expect(MockedLMStudioClientConstructor).toHaveBeenCalledWith({ baseUrl: defaultLmsUrl })
+		})
+
+		it("should transform https baseUrl to wss for LMStudioClient", async () => {
+			const httpsBaseUrl = "https://securehost:4321"
+			const wssLmsUrl = "wss://securehost:4321"
+			mockedAxios.get.mockResolvedValueOnce({ data: {} })
+			mockListLoaded.mockResolvedValueOnce([])
+
+			await getLMStudioModels(httpsBaseUrl)
+
+			expect(mockedAxios.get).toHaveBeenCalledWith(`${httpsBaseUrl}/v1/models`)
+			expect(MockedLMStudioClientConstructor).toHaveBeenCalledWith({ baseUrl: wssLmsUrl })
+		})
+
+		it("should return an empty object if lmsUrl is unparsable", async () => {
+			const unparsableBaseUrl = "http://localhost:invalid:port" // Leads to ws://localhost:invalid:port
+
+			const result = await getLMStudioModels(unparsableBaseUrl)
+
+			expect(result).toEqual({})
+			expect(mockedAxios.get).not.toHaveBeenCalled()
+			expect(MockedLMStudioClientConstructor).not.toHaveBeenCalled()
+		})
+
+		it("should return an empty object and log error if axios.get fails with a generic error", async () => {
+			const consoleErrorSpy = vi.spyOn(console, "error").mockImplementation(() => {})
+			const networkError = new Error("Network connection failed")
+			mockedAxios.get.mockRejectedValueOnce(networkError)
+
+			const result = await getLMStudioModels(baseUrl)
+
+			expect(mockedAxios.get).toHaveBeenCalledTimes(1)
+			expect(mockedAxios.get).toHaveBeenCalledWith(`${baseUrl}/v1/models`)
+			expect(MockedLMStudioClientConstructor).not.toHaveBeenCalled()
+			expect(mockListLoaded).not.toHaveBeenCalled()
+			expect(consoleErrorSpy).toHaveBeenCalledWith(
+				`Error fetching LMStudio models: ${JSON.stringify(networkError, Object.getOwnPropertyNames(networkError), 2)}`,
+			)
+			expect(result).toEqual({})
+			consoleErrorSpy.mockRestore()
+		})
+
+		it("should return an empty object and log info if axios.get fails with ECONNREFUSED", async () => {
+			const consoleInfoSpy = vi.spyOn(console, "warn").mockImplementation(() => {})
+			const econnrefusedError = new Error("Connection refused")
+			;(econnrefusedError as any).code = "ECONNREFUSED"
+			mockedAxios.get.mockRejectedValueOnce(econnrefusedError)
+
+			const result = await getLMStudioModels(baseUrl)
+
+			expect(mockedAxios.get).toHaveBeenCalledTimes(1)
+			expect(mockedAxios.get).toHaveBeenCalledWith(`${baseUrl}/v1/models`)
+			expect(MockedLMStudioClientConstructor).not.toHaveBeenCalled()
+			expect(mockListLoaded).not.toHaveBeenCalled()
+			expect(consoleInfoSpy).toHaveBeenCalledWith(`Error connecting to LMStudio at ${baseUrl}`)
+			expect(result).toEqual({})
+			consoleInfoSpy.mockRestore()
+		})
+
+		it("should return an empty object and log error if listDownloadedModels fails", async () => {
+			const consoleErrorSpy = vi.spyOn(console, "error").mockImplementation(() => {})
+			const listError = new Error("LMStudio SDK internal error")
+
+			mockedAxios.get.mockResolvedValueOnce({ data: {} })
+			mockListLoaded.mockRejectedValueOnce(listError)
+
+			const result = await getLMStudioModels(baseUrl)
+
+			expect(mockedAxios.get).toHaveBeenCalledTimes(1)
+			expect(MockedLMStudioClientConstructor).toHaveBeenCalledTimes(1)
+			expect(MockedLMStudioClientConstructor).toHaveBeenCalledWith({ baseUrl: lmsUrl })
+			expect(mockListLoaded).toHaveBeenCalledTimes(1)
+			expect(consoleErrorSpy).toHaveBeenCalledWith(
+				`Error fetching LMStudio models: ${JSON.stringify(listError, Object.getOwnPropertyNames(listError), 2)}`,
+			)
+			expect(result).toEqual({})
+			consoleErrorSpy.mockRestore()
+		})
+	})
+})