Skip to content

Use Claude Code with your ChatGPT, Kimi, Cursor or Grok subscription via a local Anthropic-compatible proxy

Rust

kimi: always log finish field in upstream chunk debug output

Raine Virtaad4c6af258ce
diff --git a/src/providers/kimi/translate/reducer.ts b/src/providers/kimi/translate/reducer.ts
index 6279f7a..a249ed5 100644
--- a/src/providers/kimi/translate/reducer.ts
+++ b/src/providers/kimi/translate/reducer.ts
@@ -67,9 +67,14 @@ interface ToolSlot {
* of typed, downstream-agnostic ReducerEvents consumed by both the
* streaming and non-streaming frontends.
*/
+export interface ReducerStats {
+ chunkCount: number
+}
+
export async function* reduceUpstream(
upstream: ReadableStream<Uint8Array>,
log: Logger,
+ stats?: ReducerStats,
): AsyncGenerator<ReducerEvent> {
let nextBlockIndex = 0
let thinkingIndex: number | undefined
@@ -114,15 +119,18 @@ export async function* reduceUpstream(
}
if (VERBOSE) {
- const d = chunk.choices?.[0]?.delta
+ const choice = chunk.choices?.[0]
+ const d = choice?.delta
log.debug("upstream chunk", {
hasReasoning: typeof d?.reasoning_content === "string",
hasContent: typeof d?.content === "string" && d.content.length > 0,
toolCalls: d?.tool_calls?.length,
- finish: chunk.choices?.[0]?.finish_reason,
+ finish: choice?.finish_reason ?? null,
})
}
+ if (stats) stats.chunkCount++
+
if (chunk.error) {
throw new UpstreamStreamError("failed", chunk.error.message || "Upstream error")
}