Add vLLM v0.18.1 source tree with KV transfer abort fix
third_party/vllm/ now tracked in git for direct patch management.
Based on vLLM v0.18.1 release with one patch applied:
vllm/v1/core/sched/scheduler.py:
Replace fatal assert with graceful skip when KV transfer callback
arrives for an already-aborted request during PD disaggregated serving.
Future vLLM modifications should be made directly in third_party/vllm/
and committed normally. The patches/ directory is kept as documentation
of what changed from upstream.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
BIN
third_party/vllm/docs/assets/design/paged_attention/k_vecs.png
vendored
Normal file
|
After Width: | Height: | Size: 27 KiB |
BIN
third_party/vllm/docs/assets/design/paged_attention/key.png
vendored
Normal file
|
After Width: | Height: | Size: 109 KiB |
BIN
third_party/vllm/docs/assets/design/paged_attention/logits_vec.png
vendored
Normal file
|
After Width: | Height: | Size: 17 KiB |
BIN
third_party/vllm/docs/assets/design/paged_attention/q_vecs.png
vendored
Normal file
|
After Width: | Height: | Size: 41 KiB |
BIN
third_party/vllm/docs/assets/design/paged_attention/query.png
vendored
Normal file
|
After Width: | Height: | Size: 32 KiB |
BIN
third_party/vllm/docs/assets/design/paged_attention/v_vec.png
vendored
Normal file
|
After Width: | Height: | Size: 42 KiB |
BIN
third_party/vllm/docs/assets/design/paged_attention/value.png
vendored
Normal file
|
After Width: | Height: | Size: 167 KiB |