Releases: gpustack/gpustack-higress-plugin
Releases · gpustack/gpustack-higress-plugin
release-20260224-172837
fix: the non-streaming requests are judged as streaming requests As the content-length header is removed by other plugin, use the content-type header to determine the data type
release-20251230-174029
fix: append chunk even if the usage exists but it is nil
release-20251226-191631
fix: refactor the streaming logic Should consider the frame has incomplete json. Assumble the chunk into a valid json before check the usage because the usage key might be split into usa and ge, then we can't determine the right usage from data.
release-20251226-120212
fix: only process the target usage chunk only If a frame has multiple chunk data which usage chunk is in the middle, should ignore the other chunk.
release-20251224-224846
fix: usage doesn't set correctly in mindIE upstream If a large usage response is sent to onStreamingResponseBody function, it will be seperated into multiple frame. It needs to be assembled into a valid json before modifying the usage data.
release-20251218-225148
fix: missing data: prefix for non usage chunk
release-20251122-232059
fix: TPOT calculates more accurately
release-20251119-154204
fix: compatible with non-streaming request
release-20251111-153636
fix: only includes chat/completions and completions paths
release-20251111-013025
fix: tpot needs to be stored in float64 format and retain 2 decimal p…