ton/pi_harness
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge remote-tracking branch 'origin/main' into model-registry

Browse Source
This commit is contained in:
Mario Zechner
2026-06-10 18:49:18 +02:00
parent f63095cfff 0ab2aa86af
commit ee276e4db5
99 changed files with 3278 additions and 796 deletions
Download Patch File Download Diff File Expand all files Collapse all files
packages/ai/src/image-models.generated.ts
+30
View File
@@ -440,6 +440,36 @@ export const IMAGE_MODELS = {
cacheWrite: 0,
},
} satisfies ImagesModel<"openrouter-images">,
"sourceful/riverflow-v2.5-fast": {
id: "sourceful/riverflow-v2.5-fast",
name: "Sourceful: Riverflow V2.5 Fast",
api: "openrouter-images",
provider: "openrouter",
baseUrl: "https://openrouter.ai/api/v1",
input: ["text", "image"],
output: ["image"],
cost: {
input: 0,
output: 0,
cacheRead: 0,
cacheWrite: 0,
},
} satisfies ImagesModel<"openrouter-images">,
"sourceful/riverflow-v2.5-pro": {
id: "sourceful/riverflow-v2.5-pro",
name: "Sourceful: Riverflow V2.5 Pro",
api: "openrouter-images",
provider: "openrouter",
baseUrl: "https://openrouter.ai/api/v1",
input: ["text", "image"],
output: ["image"],
cost: {
input: 0,
output: 0,
cacheRead: 0,
cacheWrite: 0,
},
} satisfies ImagesModel<"openrouter-images">,
"x-ai/grok-imagine-image-quality": {
id: "x-ai/grok-imagine-image-quality",
name: "xAI: Grok Imagine Image Quality",
packages/ai/src/models.generated.ts
+409 -189
View File
File diff suppressed because it is too large Load Diff
packages/ai/src/providers/amazon-bedrock.ts
+26 -8
View File
@@ -143,10 +143,13 @@ export const streamBedrock: StreamFunction<"bedrock-converse-stream", BedrockOpt
// in Node.js/Bun environment only
if (typeof process !== "undefined" && (process.versions?.node || process.versions?.bun)) {
// Region resolution: explicit option > env vars > SDK default chain.
// When AWS_PROFILE is set, we leave region undefined so the SDK can
// resovle it from aws profile configs. Otherwise fall back to us-east-1.
if (configuredRegion) {
// Region resolution: ARN-embedded > explicit option > env vars > SDK default chain.
// When the model ID is an inference profile ARN, extract the region from it.
// This avoids conflicts with AWS_REGION set for other services.
const arnRegionMatch = model.id.match(/^arn:aws(?:-[a-z0-9-]+)?:bedrock:([a-z0-9-]+):/);
if (arnRegionMatch) {
config.region = arnRegionMatch[1];
} else if (configuredRegion) {
config.region = configuredRegion;
} else if (endpointRegion && useExplicitEndpoint) {
config.region = endpointRegion;
@@ -287,6 +290,13 @@ const BEDROCK_ERROR_PREFIXES: Record<string, string> = {
ServiceUnavailableException: "Service unavailable",
};
/**
* Some models reject the account/profile's configured Bedrock data retention mode
* (e.g. "data retention mode 'default' is not available for this model"). Point
* users at the AWS docs explaining how to configure a supported mode.
*/
const BEDROCK_DATA_RETENTION_DOCS_URL = "https://docs.aws.amazon.com/bedrock/latest/userguide/data-retention.html";
/**
* Format a Bedrock error with a human-readable prefix.
* AWS SDK exceptions (both from `client.send()` and from stream event items)
@@ -296,11 +306,14 @@ const BEDROCK_ERROR_PREFIXES: Record<string, string> = {
*/
function formatBedrockError(error: unknown): string {
const message = error instanceof Error ? error.message : JSON.stringify(error);
const dataRetentionHint = /data retention mode/i.test(message)
? ` See ${BEDROCK_DATA_RETENTION_DOCS_URL} for supported data retention modes.`
: "";
if (error instanceof BedrockRuntimeServiceException) {
const prefix = BEDROCK_ERROR_PREFIXES[error.name] ?? error.name;
return `${prefix}: ${message}`;
return `${prefix}: ${message}${dataRetentionHint}`;
}
return message;
return `${message}${dataRetentionHint}`;
}
/**
@@ -525,13 +538,18 @@ function getModelMatchCandidates(modelId: string, modelName?: string): string[]
function supportsAdaptiveThinking(modelId: string, modelName?: string): boolean {
const candidates = getModelMatchCandidates(modelId, modelName);
return candidates.some(
(s) => s.includes("opus-4-6") || s.includes("opus-4-7") || s.includes("opus-4-8") || s.includes("sonnet-4-6"),
(s) =>
s.includes("opus-4-6") ||
s.includes("opus-4-7") ||
s.includes("opus-4-8") ||
s.includes("sonnet-4-6") ||
s.includes("fable-5"),
);
}
function supportsNativeXhighEffort(model: Model<"bedrock-converse-stream">): boolean {
const candidates = getModelMatchCandidates(model.id, model.name);
return candidates.some((s) => s.includes("opus-4-7") || s.includes("opus-4-8"));
return candidates.some((s) => s.includes("opus-4-7") || s.includes("opus-4-8") || s.includes("fable-5"));
}
function mapThinkingLevelToEffort(
packages/ai/src/providers/anthropic.ts
+3 -3
View File
@@ -200,7 +200,7 @@ export interface AnthropicOptions extends StreamOptions {
* Effort level for adaptive thinking models.
* Controls how much thinking Claude allocates:
* - "max": Always thinks with no constraints (Opus 4.6 only)
* - "xhigh": Highest reasoning level (Opus 4.7)
* - "xhigh": Highest reasoning level (Opus 4.7+, Fable 5)
* - "high": Always thinks, deep reasoning
* - "medium": Moderate thinking, may skip for simple queries
* - "low": Minimal thinking, skips for simple tasks
@@ -711,7 +711,7 @@ export const streamAnthropic: StreamFunction<"anthropic-messages", AnthropicOpti
/**
* Map ThinkingLevel to Anthropic effort levels for adaptive thinking.
* Note: effort "max" is only valid on Opus 4.6, while Opus 4.7 supports "xhigh".
* Note: effort "max" is only valid on Opus 4.6, while Opus 4.7+ and Fable 5 support "xhigh".
*/
function mapThinkingLevelToEffort(
model: Model<"anthropic-messages">,
@@ -972,7 +972,7 @@ function buildParams(
display,
};
}
} else if (options?.thinkingEnabled === false) {
} else if (options?.thinkingEnabled === false && model.thinkingLevelMap?.off !== null) {
params.thinking = { type: "disabled" };
}
}
packages/ai/src/providers/azure-openai-responses.ts
+1
View File
@@ -256,6 +256,7 @@ function buildParams(
input: messages,
stream: true,
prompt_cache_key: clampOpenAIPromptCacheKey(options?.sessionId),
store: false,
};
if (options?.maxTokens) {
packages/ai/src/providers/openai-completions.ts
+2 -1
View File
@@ -554,7 +554,8 @@ function buildParams(
}
if (compat.thinkingFormat === "zai" && model.reasoning) {
(params as any).enable_thinking = !!options?.reasoningEffort;
const zaiParams = params as typeof params & { thinking?: { type: "enabled" | "disabled" } };
zaiParams.thinking = { type: options?.reasoningEffort ? "enabled" : "disabled" };
} else if (compat.thinkingFormat === "qwen" && model.reasoning) {
(params as any).enable_thinking = !!options?.reasoningEffort;
} else if (compat.thinkingFormat === "qwen-chat-template" && model.reasoning) {
packages/ai/src/types.ts
+1 -1
View File
@@ -426,7 +426,7 @@ export interface OpenAICompletionsCompat {
requiresThinkingAsText?: boolean;
/** Whether all replayed assistant messages must include an empty reasoning_content field when reasoning is enabled. Default: auto-detected from URL. */
requiresReasoningContentOnAssistantMessages?: boolean;
/** Format for reasoning/thinking parameter. "openai" uses reasoning_effort, "openrouter" uses reasoning: { effort }, "deepseek" uses thinking: { type } plus reasoning_effort when supported, "together" uses reasoning: { enabled } plus reasoning_effort when supported, "zai" uses top-level enable_thinking: boolean, "qwen" uses top-level enable_thinking: boolean, "qwen-chat-template" uses chat_template_kwargs.enable_thinking, "string-thinking" uses top-level thinking: string, and "ant-ling" uses reasoning: { effort } only when the mapped effort is non-null. Default: "openai". */
/** Format for reasoning/thinking parameter. "openai" uses reasoning_effort, "openrouter" uses reasoning: { effort }, "deepseek" uses thinking: { type } plus reasoning_effort when supported, "together" uses reasoning: { enabled } plus reasoning_effort when supported, "zai" uses thinking: { type }, "qwen" uses top-level enable_thinking: boolean, "qwen-chat-template" uses chat_template_kwargs.enable_thinking, "string-thinking" uses top-level thinking: string, and "ant-ling" uses reasoning: { effort } only when the mapped effort is non-null. Default: "openai". */
thinkingFormat?:
| "openai"
| "openrouter"