Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
126 changes: 26 additions & 100 deletions Cargo.lock

Some generated files are not rendered by default. Learn more about how customized files appear on GitHub.

10 changes: 10 additions & 0 deletions docs/how-to/serve-a-model-on-spark.md
Original file line number Diff line number Diff line change
Expand Up @@ -119,6 +119,16 @@ model-identity probe pass, OpenAI-compatible routes are at
routes are under `/anthropic/<instance>/v1`. Warming or recovering
generations return protocol-native `503` with `Retry-After`.

For translated Responses requests, an engine can occasionally stop after
reasoning without producing an answer or tool call. The gateway makes at most
one assistant-prefill continuation, retaining the request history, tools and
sampling settings. The engine's chat template closes the retained reasoning
block so generation can continue into the answer. Both attempts share the
remaining output budget, and reported usage includes both generations. This
does not reset or compact the coding session. Unsupported continuations and
repeated reasoning-only stops fail explicitly; malformed tool arguments and
length-limited generations are not retried by this recovery path.

For Ornith, reasoning streams independently from final answer text as OpenAI
`reasoning_content`, Responses reasoning-summary events, or Anthropic thinking
blocks. `sparkplane ... launch codex` advertises that capability in its private
Expand Down
Loading
Loading