mirror of
https://github.com/getcompanion-ai/co-mono.git
synced 2026-04-21 08:02:11 +00:00
refactor(ai): Update API to support partial results on abort
- Anthropic, Google, and OpenAI Responses providers now return partial results when aborted - Restructured streaming to accumulate content blocks incrementally - Prevents submission of thinking/toolCall blocks from aborted completions in multi-turn conversations - Makes UI development easier by providing partial content even when requests are interrupted
This commit is contained in:
Mario Zechner
parent
5d5cd7955b
commit
bf1f410c2b
4 changed files with 244 additions and 280 deletions
|
|
@ -22,7 +22,6 @@ import type {
|
|||
TextContent,
|
||||
Tool,
|
||||
ToolCall,
|
||||
Usage,
|
||||
} from "../types.js";
|
||||
|
||||
export interface OpenAIResponsesLLMOptions extends LLMOptions {
|
||||
|
|
@ -52,6 +51,20 @@ export class OpenAIResponsesLLM implements LLM<OpenAIResponsesLLMOptions> {
|
|||
}
|
||||
|
||||
async complete(request: Context, options?: OpenAIResponsesLLMOptions): Promise<AssistantMessage> {
|
||||
const output: AssistantMessage = {
|
||||
role: "assistant",
|
||||
content: [],
|
||||
provider: this.modelInfo.provider,
|
||||
model: this.modelInfo.id,
|
||||
usage: {
|
||||
input: 0,
|
||||
output: 0,
|
||||
cacheRead: 0,
|
||||
cacheWrite: 0,
|
||||
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
|
||||
},
|
||||
stopReason: "stop",
|
||||
};
|
||||
try {
|
||||
const input = this.convertToInput(request.messages, request.systemPrompt);
|
||||
|
||||
|
|
@ -88,17 +101,8 @@ export class OpenAIResponsesLLM implements LLM<OpenAIResponsesLLMOptions> {
|
|||
|
||||
options?.onEvent?.({ type: "start", model: this.modelInfo.id, provider: this.modelInfo.provider });
|
||||
|
||||
const outputItems: (ResponseReasoningItem | ResponseOutputMessage | ResponseFunctionToolCall)[] = []; // any for function_call items
|
||||
let currentTextAccum = ""; // For delta accumulation
|
||||
let currentThinkingAccum = ""; // For delta accumulation
|
||||
let usage: Usage = {
|
||||
input: 0,
|
||||
output: 0,
|
||||
cacheRead: 0,
|
||||
cacheWrite: 0,
|
||||
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
|
||||
};
|
||||
let stopReason: StopReason = "stop";
|
||||
const outputItems: (ResponseReasoningItem | ResponseOutputMessage | ResponseFunctionToolCall)[] = [];
|
||||
let currentItem: ResponseReasoningItem | ResponseOutputMessage | ResponseFunctionToolCall | null = null;
|
||||
|
||||
for await (const event of stream) {
|
||||
// Handle output item start
|
||||
|
|
@ -106,47 +110,91 @@ export class OpenAIResponsesLLM implements LLM<OpenAIResponsesLLMOptions> {
|
|||
const item = event.item;
|
||||
if (item.type === "reasoning") {
|
||||
options?.onEvent?.({ type: "thinking_start" });
|
||||
currentThinkingAccum = "";
|
||||
outputItems.push(item);
|
||||
currentItem = item;
|
||||
} else if (item.type === "message") {
|
||||
options?.onEvent?.({ type: "text_start" });
|
||||
currentTextAccum = "";
|
||||
outputItems.push(item);
|
||||
currentItem = item;
|
||||
}
|
||||
}
|
||||
// Handle reasoning summary deltas
|
||||
else if (event.type === "response.reasoning_summary_text.delta") {
|
||||
const delta = event.delta;
|
||||
currentThinkingAccum += delta;
|
||||
options?.onEvent?.({ type: "thinking_delta", content: currentThinkingAccum, delta });
|
||||
else if (event.type === "response.reasoning_summary_part.added") {
|
||||
if (currentItem && currentItem.type === "reasoning") {
|
||||
currentItem.summary = currentItem.summary || [];
|
||||
currentItem.summary.push(event.part);
|
||||
}
|
||||
} else if (event.type === "response.reasoning_summary_text.delta") {
|
||||
if (currentItem && currentItem.type === "reasoning") {
|
||||
currentItem.summary = currentItem.summary || [];
|
||||
const lastPart = currentItem.summary[currentItem.summary.length - 1];
|
||||
if (lastPart) {
|
||||
lastPart.text += event.delta;
|
||||
options?.onEvent?.({
|
||||
type: "thinking_delta",
|
||||
content: currentItem.summary.join("\n\n"),
|
||||
delta: event.delta,
|
||||
});
|
||||
}
|
||||
}
|
||||
}
|
||||
// Add a new line between summary parts (hack...)
|
||||
else if (event.type === "response.reasoning_summary_part.done") {
|
||||
currentThinkingAccum += "\n\n";
|
||||
options?.onEvent?.({ type: "thinking_delta", content: currentThinkingAccum, delta: "\n\n" });
|
||||
if (currentItem && currentItem.type === "reasoning") {
|
||||
options?.onEvent?.({
|
||||
type: "thinking_delta",
|
||||
content: currentItem.summary.join("\n\n"),
|
||||
delta: "\n\n",
|
||||
});
|
||||
}
|
||||
}
|
||||
// Handle text output deltas
|
||||
else if (event.type === "response.output_text.delta") {
|
||||
const delta = event.delta;
|
||||
currentTextAccum += delta;
|
||||
options?.onEvent?.({ type: "text_delta", content: currentTextAccum, delta });
|
||||
}
|
||||
// Handle refusal output deltas
|
||||
else if (event.type === "response.refusal.delta") {
|
||||
const delta = event.delta;
|
||||
currentTextAccum += delta;
|
||||
options?.onEvent?.({ type: "text_delta", content: currentTextAccum, delta });
|
||||
else if (event.type === "response.content_part.added") {
|
||||
if (currentItem && currentItem.type === "message") {
|
||||
currentItem.content = currentItem.content || [];
|
||||
currentItem.content.push(event.part);
|
||||
}
|
||||
} else if (event.type === "response.output_text.delta") {
|
||||
if (currentItem && currentItem.type === "message") {
|
||||
const lastPart = currentItem.content[currentItem.content.length - 1];
|
||||
if (lastPart && lastPart.type === "output_text") {
|
||||
lastPart.text += event.delta;
|
||||
options?.onEvent?.({
|
||||
type: "text_delta",
|
||||
content: currentItem.content
|
||||
.map((c) => (c.type === "output_text" ? c.text : c.refusal))
|
||||
.join(""),
|
||||
delta: event.delta,
|
||||
});
|
||||
}
|
||||
}
|
||||
} else if (event.type === "response.refusal.delta") {
|
||||
if (currentItem && currentItem.type === "message") {
|
||||
const lastPart = currentItem.content[currentItem.content.length - 1];
|
||||
if (lastPart && lastPart.type === "refusal") {
|
||||
lastPart.refusal += event.delta;
|
||||
options?.onEvent?.({
|
||||
type: "text_delta",
|
||||
content: currentItem.content
|
||||
.map((c) => (c.type === "output_text" ? c.text : c.refusal))
|
||||
.join(""),
|
||||
delta: event.delta,
|
||||
});
|
||||
}
|
||||
}
|
||||
}
|
||||
// Handle output item completion
|
||||
else if (event.type === "response.output_item.done") {
|
||||
const item = event.item;
|
||||
|
||||
if (item.type === "reasoning") {
|
||||
outputItems[outputItems.length - 1] = item; // Update with final item
|
||||
const thinkingContent = item.summary?.map((s: any) => s.text).join("\n\n") || "";
|
||||
options?.onEvent?.({ type: "thinking_end", content: thinkingContent });
|
||||
outputItems.push(item);
|
||||
} else if (item.type === "message") {
|
||||
outputItems[outputItems.length - 1] = item; // Update with final item
|
||||
const textContent = item.content.map((c) => (c.type === "output_text" ? c.text : c.refusal)).join("");
|
||||
options?.onEvent?.({ type: "text_end", content: textContent });
|
||||
outputItems.push(item);
|
||||
} else if (item.type === "function_call") {
|
||||
const toolCall: ToolCall = {
|
||||
type: "toolCall",
|
||||
|
|
@ -162,7 +210,7 @@ export class OpenAIResponsesLLM implements LLM<OpenAIResponsesLLMOptions> {
|
|||
else if (event.type === "response.completed") {
|
||||
const response = event.response;
|
||||
if (response?.usage) {
|
||||
usage = {
|
||||
output.usage = {
|
||||
input: response.usage.input_tokens || 0,
|
||||
output: response.usage.output_tokens || 0,
|
||||
cacheRead: response.usage.input_tokens_details?.cached_tokens || 0,
|
||||
|
|
@ -170,60 +218,43 @@ export class OpenAIResponsesLLM implements LLM<OpenAIResponsesLLMOptions> {
|
|||
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
|
||||
};
|
||||
}
|
||||
|
||||
calculateCost(this.modelInfo, output.usage);
|
||||
// Map status to stop reason
|
||||
stopReason = this.mapStopReason(response?.status);
|
||||
output.stopReason = this.mapStopReason(response?.status);
|
||||
if (outputItems.some((b) => b.type === "function_call") && output.stopReason === "stop") {
|
||||
output.stopReason = "toolUse";
|
||||
}
|
||||
}
|
||||
// Handle errors
|
||||
else if (event.type === "error") {
|
||||
const errorOutput = {
|
||||
role: "assistant",
|
||||
content: [],
|
||||
provider: this.modelInfo.provider,
|
||||
model: this.modelInfo.id,
|
||||
usage,
|
||||
stopReason: "error",
|
||||
error: `Code ${event.code}: ${event.message}` || "Unknown error",
|
||||
} satisfies AssistantMessage;
|
||||
options?.onEvent?.({ type: "error", error: errorOutput.error || "Unknown error" });
|
||||
return errorOutput;
|
||||
output.stopReason = "error";
|
||||
output.error = `Code ${event.code}: ${event.message}` || "Unknown error";
|
||||
options?.onEvent?.({ type: "error", error: output.error });
|
||||
return output;
|
||||
} else if (event.type === "response.failed") {
|
||||
const errorOutput = {
|
||||
role: "assistant",
|
||||
content: [],
|
||||
provider: this.modelInfo.provider,
|
||||
model: this.modelInfo.id,
|
||||
usage,
|
||||
stopReason: "error",
|
||||
error: "Unknown error",
|
||||
} satisfies AssistantMessage;
|
||||
options?.onEvent?.({ type: "error", error: errorOutput.error || "Unknown error" });
|
||||
return errorOutput;
|
||||
output.stopReason = "error";
|
||||
output.error = "Unknown error";
|
||||
options?.onEvent?.({ type: "error", error: output.error });
|
||||
return output;
|
||||
}
|
||||
}
|
||||
|
||||
if (options?.signal?.aborted) {
|
||||
throw new Error("Request was aborted");
|
||||
}
|
||||
|
||||
// Convert output items to blocks
|
||||
const blocks: AssistantMessage["content"] = [];
|
||||
|
||||
for (const item of outputItems) {
|
||||
if (item.type === "reasoning") {
|
||||
blocks.push({
|
||||
output.content.push({
|
||||
type: "thinking",
|
||||
thinking: item.summary?.map((s: any) => s.text).join("\n\n") || "",
|
||||
thinkingSignature: JSON.stringify(item), // Full item for resubmission
|
||||
});
|
||||
} else if (item.type === "message") {
|
||||
blocks.push({
|
||||
output.content.push({
|
||||
type: "text",
|
||||
text: item.content.map((c) => (c.type === "output_text" ? c.text : c.refusal)).join(""),
|
||||
textSignature: item.id, // ID for resubmission
|
||||
});
|
||||
} else if (item.type === "function_call") {
|
||||
blocks.push({
|
||||
output.content.push({
|
||||
type: "toolCall",
|
||||
id: item.call_id + "|" + item.id,
|
||||
name: item.name,
|
||||
|
|
@ -232,40 +263,16 @@ export class OpenAIResponsesLLM implements LLM<OpenAIResponsesLLMOptions> {
|
|||
}
|
||||
}
|
||||
|
||||
// Check if we have tool calls for stop reason
|
||||
if (blocks.some((b) => b.type === "toolCall") && stopReason === "stop") {
|
||||
stopReason = "toolUse";
|
||||
if (options?.signal?.aborted) {
|
||||
throw new Error("Request was aborted");
|
||||
}
|
||||
|
||||
calculateCost(this.modelInfo, usage);
|
||||
|
||||
const output = {
|
||||
role: "assistant",
|
||||
content: blocks,
|
||||
provider: this.modelInfo.provider,
|
||||
model: this.modelInfo.id,
|
||||
usage,
|
||||
stopReason,
|
||||
} satisfies AssistantMessage;
|
||||
options?.onEvent?.({ type: "done", reason: output.stopReason, message: output });
|
||||
return output;
|
||||
} catch (error) {
|
||||
const output = {
|
||||
role: "assistant",
|
||||
content: [],
|
||||
provider: this.modelInfo.provider,
|
||||
model: this.modelInfo.id,
|
||||
usage: {
|
||||
input: 0,
|
||||
output: 0,
|
||||
cacheRead: 0,
|
||||
cacheWrite: 0,
|
||||
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
|
||||
},
|
||||
stopReason: "error",
|
||||
error: error instanceof Error ? error.message : String(error),
|
||||
} satisfies AssistantMessage;
|
||||
options?.onEvent?.({ type: "error", error: output.error || "Unknown error" });
|
||||
output.stopReason = "error";
|
||||
output.error = error instanceof Error ? error.message : JSON.stringify(error);
|
||||
options?.onEvent?.({ type: "error", error: output.error });
|
||||
return output;
|
||||
}
|
||||
}
|
||||
|
|
@ -318,7 +325,8 @@ export class OpenAIResponsesLLM implements LLM<OpenAIResponsesLLMOptions> {
|
|||
const output: ResponseInput = [];
|
||||
|
||||
for (const block of msg.content) {
|
||||
if (block.type === "thinking") {
|
||||
// Do not submit thinking blocks if the completion had an error (i.e. abort)
|
||||
if (block.type === "thinking" && msg.stopReason !== "error") {
|
||||
// Push the full reasoning item(s) from signature
|
||||
if (block.thinkingSignature) {
|
||||
const reasoningItem = JSON.parse(block.thinkingSignature);
|
||||
|
|
@ -333,7 +341,8 @@ export class OpenAIResponsesLLM implements LLM<OpenAIResponsesLLMOptions> {
|
|||
status: "completed",
|
||||
id: textBlock.textSignature || "msg_" + Math.random().toString(36).substring(2, 15),
|
||||
} satisfies ResponseOutputMessage);
|
||||
} else if (block.type === "toolCall") {
|
||||
// Do not submit thinking blocks if the completion had an error (i.e. abort)
|
||||
} else if (block.type === "toolCall" && msg.stopReason !== "error") {
|
||||
const toolCall = block as ToolCall;
|
||||
output.push({
|
||||
type: "function_call",
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue