Releases · vllm-project/vime

What's Changed

fix: add coding_agent_rl dependencies to requirements.txt by @aoshen02 in #216
fix: complete #1985 sync — remove TIGHT_HOST_MEMORY + align test params by @aoshen02 in #218
fix: refine positioning wording in README by @aoshen02 in #221
fix(generate): remove spurious model= kwarg from AnthropicAdapter call by @aoshen02 in #219
Update README.md by @aoshen02 in #222
docs: acknowledge SkyRL/verl and list SkyRL/prime-rl as users in README by @aoshen02 in #225
Add Read the Docs docs configuration by @aoshen02 in #152
fix(test): align fanout_short assertion with #218 param changes by @aoshen02 in #226
[Doc] Fix Read the Docs build config by @aoshen02 in #234
[Doc] Fix RTD language toggle paths by @aoshen02 in #236
sync: complete slime #1920, #1967, #1985 by @aoshen02 in #232
[Doc] Support RTD zh-cn translation paths by @aoshen02 in #237
refactor(test): flatten CPU unit tests and centralize import stubs by @CalvinXKY in #238
docs: align vLLM docs with current defaults by @aoshen02 in #240
refactor(weight-sync): use vLLM native /update_weights, remove worker extension by @aoshen02 in #246
Add Buildkite CI pipeline (CPU jobs + manual-gated GPU suites) and remove PR test on GHA by @khluu in #239
[codex] update docs branding and fix Buildkite CPU tests by @aoshen02 in #249
fix(arguments): include DP in TP auto-compute default by @aoshen02 in #252
[Example] Add tau-bench multi-turn tool-use example by @CalvinXKY in #142
[codex] drop stale unit CPU CI and tests by @aoshen02 in #268
docker: upgrade base to vLLM 0.23.0, remove CUDA 13 build path by @aoshen02 in #253
refactor(vllm_engine): in-process launch + dataclass field introspection by @aoshen02 in #264
[codex] docs: fix Qwen3-4B download repo by @aoshen02 in #276
test: stub vllm.utils.system_utils in CPU unit test stubs by @aoshen02 in #284
fix(data): reuse stored multimodal_inputs in length filter by @Meihan-chen in #257
docker: rename published image from inferactinc/public to vllm/vime by @aoshen02 in #283
fix(examples/tau-bench): use RunConfig.agent_strategy in TAU_CONFIGS (slime #2101) by @CalvinXKY in #280
scripts: complete slime-exact port of most scripts except for gpt-oss 20B support by @aoshen02 in #260
[Doc] Fix broken Qwen3-4B example link in rollout_buffer README by @ajinkyajawale14499 in #295
sync(slime #2014..#2125): 3-way merge [WIP — 45 conflict files need review] by @aoshen02 in #286
[Example] Add MemAgent long-context RL example (mem_agent) by @CalvinXKY in #291
fix(rollout): abort vLLM rollout via delete-type /abort_requests by @aoshen02 in #296
[Bugfix][Rollout] Wire prefix_cache_hit_rate through vLLM usage by @aoshen02 in #303
fix(docker): /abort_requests abort-all fix + cu13 image variant by @aoshen02 in #317
Initial ROCm support for vime by @indianspeedster in #273
[Doc] Fix stale Qwen3-30B-A3B example: restore FP8-inference section + rewrite multi-node by @aoshen02 in #322
[Doc] Fix H800→H100 typo in Qwen3-30B-A3B example by @aoshen02 in #325
[Doc] Add Ascend NPU platform tutorial and expand hardware support section by @CalvinXKY in #334
sync: update vime through slime #2185 by @aoshen02 in #338
revert: sync through slime #2185 (#338) by @aoshen02 in #342
[bugfix] Fix distributed update weights for pipeline parallel by @knlnguyen1802 in #329
weekly sync: update vime through slime #2185 by @aoshen02 in #343
[Bugfix][Rollout] Validate batched RM reward lengths by @BreezyB1n in #313
Support online MTP draft weight updates by @aoshen02 in #351
refactor(rollout): mirror slime router config — cache_aware default + disable circuit breaker by @aoshen02 in #350
Forward recompute flags to the bridge provider; two hybrid model fixes by @yuchenwang3 in #337
docker: bump default CU13 image to vLLM 0.25.1 by @aoshen02 in #353
[Bugfix][Rollout] Fix Geo3K VLM multi-turn rollout by @FeathBow in #341
perf: accelerate colocated and non-colocated vLLM weight updates by @aoshen02 in #340
[Bugfix][Rollout] Fix IPv6 vLLM engine and health-check URLs by @ShuoleiWang in #357
[Bugfix] Sync missing Slime runtime safeguards by @aoshen02 in #359
Add on-policy distillation example (Qwen3-8B + Qwen3-32B vLLM teacher) by @CalvinXKY in #328
ci(rocm): AMD ROCm GPU CI on Buildkite by @indianspeedster in #356
docs: document cache-aware vLLM router support by @aoshen02 in #371
[Rollout] Add vLLM encoder-prefill disaggregation by @Gianthard-cyh in #370
sync: update from Slime and upgrade to latest vLLM nightly by @aoshen02 in #386
docker: split pull-weight patch and drop upstreamed vLLM fixes by @aoshen02 in #398
docs: surface agent workflows and fix vLLM response parsing by @aoshen02 in #401
[CI/Build] Run the CPU tests CI silently skips, and guard against new gaps by @natedemoss in #400
[Training] Add Qwen3-Omni multimodal RL support by @CalvinXKY in #378
feat: DSpark speculative decoding draft model training for RL rollout acceleration by @CalvinXKY in #397
ci(rocm): rename ROCm queue to amd_mi355_vime_rl by @indianspeedster in #405
[Sync] Update to Slime v0.3.2 and vLLM nightly by @aoshen02 in #402

No results found