shumengya/sproutclaw
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: update zAI models to use anthropic API and filter empty thinking blocks in messages

Browse Source
This commit is contained in:
Anton Kuzmenko
2025-12-28 16:31:32 -08:00
parent e9c6d95e6b
commit 31cbbd211c
4 changed files with 36 additions and 34 deletions
Download Patch File Download Diff File Expand all files Collapse all files

35
packages/ai/src/models.generated.ts
View File

@@ -6978,10 +6978,9 @@ export const MODELS = {
"glm-4.5": {
id: "glm-4.5",
name: "GLM-4.5",
api: "openai-completions",
api: "anthropic-messages",
provider: "zai",
baseUrl: "https://api.z.ai/api/coding/paas/v4",
compat: {"supportsDeveloperRole":false},
baseUrl: "https://api.z.ai/api/anthropic",
reasoning: true,
input: ["text"],
cost: {
@@ -6992,14 +6991,13 @@ export const MODELS = {
},
contextWindow: 131072,
maxTokens: 98304,
} satisfies Model<"openai-completions">,
} satisfies Model<"anthropic-messages">,
"glm-4.5-air": {
id: "glm-4.5-air",
name: "GLM-4.5-Air",
api: "openai-completions",
api: "anthropic-messages",
provider: "zai",
baseUrl: "https://api.z.ai/api/coding/paas/v4",
compat: {"supportsDeveloperRole":false},
baseUrl: "https://api.z.ai/api/anthropic",
reasoning: true,
input: ["text"],
cost: {
@@ -7010,14 +7008,13 @@ export const MODELS = {
},
contextWindow: 131072,
maxTokens: 98304,
} satisfies Model<"openai-completions">,
} satisfies Model<"anthropic-messages">,
"glm-4.5-flash": {
id: "glm-4.5-flash",
name: "GLM-4.5-Flash",
api: "openai-completions",
api: "anthropic-messages",
provider: "zai",
baseUrl: "https://api.z.ai/api/coding/paas/v4",
compat: {"supportsDeveloperRole":false},
baseUrl: "https://api.z.ai/api/anthropic",
reasoning: true,
input: ["text"],
cost: {
@@ -7028,7 +7025,7 @@ export const MODELS = {
},
contextWindow: 131072,
maxTokens: 98304,
} satisfies Model<"openai-completions">,
} satisfies Model<"anthropic-messages">,
"glm-4.5v": {
id: "glm-4.5v",
name: "GLM-4.5V",
@@ -7050,10 +7047,9 @@ export const MODELS = {
"glm-4.6": {
id: "glm-4.6",
name: "GLM-4.6",
api: "openai-completions",
api: "anthropic-messages",
provider: "zai",
baseUrl: "https://api.z.ai/api/coding/paas/v4",
compat: {"supportsDeveloperRole":false},
baseUrl: "https://api.z.ai/api/anthropic",
reasoning: true,
input: ["text"],
cost: {
@@ -7064,7 +7060,7 @@ export const MODELS = {
},
contextWindow: 204800,
maxTokens: 131072,
} satisfies Model<"openai-completions">,
} satisfies Model<"anthropic-messages">,
"glm-4.6v": {
id: "glm-4.6v",
name: "GLM-4.6V",
@@ -7086,10 +7082,9 @@ export const MODELS = {
"glm-4.7": {
id: "glm-4.7",
name: "GLM-4.7",
api: "openai-completions",
api: "anthropic-messages",
provider: "zai",
baseUrl: "https://api.z.ai/api/coding/paas/v4",
compat: {"supportsDeveloperRole":false},
baseUrl: "https://api.z.ai/api/anthropic",
reasoning: true,
input: ["text"],
cost: {
@@ -7100,6 +7095,6 @@ export const MODELS = {
},
contextWindow: 204800,
maxTokens: 131072,
} satisfies Model<"openai-completions">,
} satisfies Model<"anthropic-messages">,
},
} as const;

12
packages/ai/src/providers/openai-completions.ts
View File

@@ -474,10 +474,14 @@ function convertMessages(
// Handle thinking blocks
const thinkingBlocks = msg.content.filter((b) => b.type === "thinking") as ThinkingContent[];
if (thinkingBlocks.length > 0) {
// Filter out empty thinking blocks to avoid API validation errors
const nonEmptyThinkingBlocks = thinkingBlocks.filter((b) => b.thinking && b.thinking.trim().length > 0);
if (nonEmptyThinkingBlocks.length > 0) {
if (compat.requiresThinkingAsText) {
// Convert thinking blocks to text with <thinking> delimiters
const thinkingText = thinkingBlocks.map((b) => `<thinking>\n${b.thinking}\n</thinking>`).join("\n");
const thinkingText = nonEmptyThinkingBlocks
.map((b) => `<thinking>\n${b.thinking}\n</thinking>`)
.join("\n");
const textContent = assistantMsg.content as Array<{ type: "text"; text: string }> | null;
if (textContent) {
textContent.unshift({ type: "text", text: thinkingText });
@@ -486,9 +490,9 @@ function convertMessages(
}
} else {
// Use the signature from the first thinking block if available (for llama.cpp server + gpt-oss)
const signature = thinkingBlocks[0].thinkingSignature;
const signature = nonEmptyThinkingBlocks[0].thinkingSignature;
if (signature && signature.length > 0) {
(assistantMsg as any)[signature] = thinkingBlocks.map((b) => b.thinking).join("\n");
(assistantMsg as any)[signature] = nonEmptyThinkingBlocks.map((b) => b.thinking).join("\n");
}
}
}