feat: add litellm.compress() — BM25-based prompt compression with retrieval tool - #25637

Merged

krrish-berri-2 merged 11 commits into

litellm_dev_04_13_2026_p1from

claude/zealous-black

Apr 13, 2026

Commits on Apr 13, 2026

feat: add litellm.compress() for BM25-based context compression

krrish-berri-2
and
claude
committed
051ad73
fix(compress): truncate high-scoring messages instead of fully stubbing them

krrish-berri-2
and
claude
committed
5c06474
fix(bm25): add prefix expansion so query terms match inflected doc tokens

krrish-berri-2
and
claude
committed
f40aa6a
test: add routing correctness test and eval harness for litellm.compress()

krrish-berri-2
and
claude
committed
c3a3de8
feat(eval): add SWE-bench Lite compression eval harness

krrish-berri-2
and
claude
committed
54c2dc6
fix(eval): robust dataset loading + sys.path fix for worktree imports

krrish-berri-2
and
claude
committed
3ae571a
improve compression quality: line-based truncation, multi-message budget, 70% default target

krrish-berri-2
and
claude
committed
6fae702
docs: add SWE-bench performance results to compress() docs

krrish-berri-2
and
claude
committed
b0e4d9c
fix(eval): use tolerance-based hunk overlap metric

krrish-berri-2
and
claude
committed
6d96ce8
feat: add compression_interception callback for LiteLLM Proxy

krrish-berri-2
and
claude
committed
72bd5cb
Revert "feat: add compression_interception callback for LiteLLM Proxy"
krrish-berri-2
committed
6e22974