v0.11.0-EPD

What's Changed

• [Bugfix] update mm hash key with request id by @Bounty-hunter in #181
• [feature]async schedule for ec connector by @Shirley125 in #182
• [bugfix]fix load failed bug by @Shirley125 in #189
• [Feature] Add encode time by @LJH-LBJ in #192
• adapt ascend direct transport by @Shirley125 in #198
• Yuanrong-datasystem adapt async schedule. by @yangsonglin13 in #201

New Contributors

• @LJH-LBJ made their first contribution in #192

Full Changelog: v0.11.0rc4-EPD...v0.11.0-EPD

AFD-POC

Full Changelog: https://github.com/JiusiServe/vllm/commits/AFD-POC

v0.11.0rc4-EPD

What's Changed

• [BugFix] Encode cache miss by @Bounty-hunter in #170
• Datasystem store: add fast transfer support by @yangsonglin13 in #168
• [BugFix] no load weight in tarsier by @Bounty-hunter in #174
• add serial methods to encode/decode tensor and mm_features by @Shirley125 in #172
• [BugFix] qwen3 no load by @Bounty-hunter in #179

Full Changelog: v0.11.0rc3-EPD...v0.11.0rc4-EPD

v0.11.0rc3-EPD

Full Changelog: https://github.com/JiusiServe/vllm/commits/v0.11.0rc3-EPD

What's Changed

• [Performance] skip embedding check if it have computed by @Bounty-hunter in #146
• [bugfix]setup mooncake store by config by @Shirley125 in #131
• Implement datasystem store to ECMooncakeConnector. by @yangsonglin13 in #151

New Contributors

• @yangsonglin13 made their first contribution in #151

Full Changelog: v0.11.0rc2-EPD...v0.11.0rc3-EPD

v0.11.0rc2-EPD

What's Changed

• [BugFix]: embedding cache bug by @Bounty-hunter in #118
• [Bugfix] Change the default value for ec_transfer_config as None by @JohnLiu97Huawei in #121
• [cherry-pick] no load llm weight in encoder of Qwen-2.5-VL by @Bounty-hunter in #128
• [Bugfix][cherry-pick] "Can not write request body for" by @Bounty-hunter in #137

Full Changelog: v0.11.0rc1-EPD...v0.11.0rc2-EPD

v0.11.0rc1-EPD

What's Changed

• [Feature]: EPD base code by @JohnLiu97Huawei in #106
• [Feature]Mooncake store ECConnector by @Shirley125 in #110

New Contributors

• @JohnLiu97Huawei made their first contribution in #106
• @Shirley125 made their first contribution in #110

Full Changelog: https://github.com/hsliuustc0106/vllm/commits/v0.11.0rc1-EPD

v0.9.1rc1-EPD

What's Changed

• [Feature]: support schedule & health check for zmq by @Bounty-hunter in #73
• [Feature] Add health check threshold by @Bounty-hunter in #68
• [Bug Fix] Reject repeat request_id & param type check & fix hang for server exception by @Bounty-hunter in #89
• [Bugfix]Return single image data when image array length is 1 in _decode_mm_data by @amy-why-3459 in #91
• 【bugfix】Support individual health check calls by @amy-why-3459 in #93

Full Changelog: v0.9.1-EPD...v0.9.1rc1-EPD

v0.9.1-EPD

What's Changed

• [Feature][EPD] Version 0.9.1 supports EPD by @amy-why-3459 in #15
• [Bugfix] "Can not write request body" by @Bounty-hunter in #31
• [BugFix] Exception: Invalid prefix encountered by @Bounty-hunter in #34
• [BugFix] Rm unused embedding file by @Bounty-hunter in #35
• cherry-pick 24519 by @amy-why-3459 in #39
• [BugFix] Not load llm weights in Encoder for llama by @Bounty-hunter in #50
• [Feature]: Support extension of scheduling strategy in proxy by @Bounty-hunter in #49
• [Bug Fix] error when text request by @Bounty-hunter in #52
• [Feature][EPD]Bypass vLLM serving framework by using zmq & its python API by @wuhang2014 in #54
• [Bug Fix] oom: pd repeate to compute mm encoder by @Bounty-hunter in #58

New Contributors

• @amy-why-3459 made their first contribution in #15
• @Bounty-hunter made their first contribution in #31
• @wuhang2014 made their first contribution in #54

Full Changelog: https://github.com/hsliuustc0106/vllm/commits/v0.9.1-EPD