How can I create machine-readable output in a specific order?
I created vLLM serverless and enabled GUIDED_DECODING_BACKEND and access it with API with
extra_body={
"guided_json": {
it does enforce the json output, but the order of properties is not enforced, despite both backends claiming to enforce this (LMFE_STRICT_JSON_FIELD_ORDER = "true").
What's the easiest way to do inference with LoRA confined to structured outputs in a specific order from a small LLM?
Doesn't need to be json. Just output reasoning first, then the answer components, in a way that each component can be 100% extracted without any extra garbage included?
extra_body={
"guided_json": {
it does enforce the json output, but the order of properties is not enforced, despite both backends claiming to enforce this (LMFE_STRICT_JSON_FIELD_ORDER = "true").
What's the easiest way to do inference with LoRA confined to structured outputs in a specific order from a small LLM?
Doesn't need to be json. Just output reasoning first, then the answer components, in a way that each component can be 100% extracted without any extra garbage included?
