Supportd distributed ray for vllm #453
Open
+238
−57
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Contributor
JingofXin
commented
Mar 3, 2025
JingofXin
commented
wzh1994
reviewed
Mar 3, 2025
lazyllm/components/deploy/vllm.py
Outdated
| return {"decode_unicode": False, "delimiter": b"\0"} | ||
| if Vllm.vllm_version is None: | ||
| Vllm.vllm_version = importlib.import_module('vllm').__version__ | ||
| if Vllm.vllm_version <= "0.5.0": |
There was a problem hiding this comment.
这个判断不对，比如'0.5.0' < '0.11.0'，结果会是False
lazyllm/components/deploy/ray.py
Outdated
| lazyllm.config.add('num_gpus_per_node', int, 8, 'NUM_GPUS_PER_NODE') | ||
|
|
||
| def reallocate_launcher(launcher): | ||
| if not isinstance(launcher, (launchers.ScoLauncher, launchers.SlurmLauncher, launchers.RemoteLauncher)): |
lazyllm/components/deploy/ray.py
Outdated
| f"limit{(lazyllm.config['num_gpus_per_node'])}. Please check the actual " | ||
| 'number of GPUs in a single node and set the environment variable: LAZYLLM_NUM_GPUS_PER_NODE. ' | ||
| 'Now LazyLLM will reconfigure the number of nodes and GPUs') | ||
| nnode = nnode if nnode > 0 else 1 # avoid 0 |
| if not isinstance(launcher, (launchers.ScoLauncher, launchers.SlurmLauncher, launchers.RemoteLauncher)): | ||
| return [], launcher | ||
| nnode = launcher.nnode | ||
| ngpus = launcher.ngpus |
There was a problem hiding this comment.
这里在多机的情况下，ngpus定义是一共多少个，还是每个node多少个？
如果是每个机器，是不是应该叫ngpus_per_node，然后超了直接报错，而不是给他重算一下加机器？
lazyllm/components/deploy/vllm.py
Outdated
| master_ip = '' | ||
| for launcher in self.launcher_list: | ||
| m = Distributed(launcher=launcher, master_ip=master_ip) | ||
| m() |
There was a problem hiding this comment.
这里在算cmd的时候就已经把任务启动是有问题的，要一起等到推理任务启动时候再启动
wzh1994
reviewed
Mar 5, 2025
lazyllm/components/deploy/vllm.py
Outdated
| @@ -53,7 +53,13 @@ def __init__(self, trust_remote_code=True, launcher=launchers.remote(ngpus=1), s | |||
| self.temp_folder = make_log_dir(log_path, 'vllm') if log_path else None | |||
| if self.launcher_list: | |||
| ray_launcher = [Distributed(launcher=launcher) for launcher in self.launcher_list] | |||
| self._prepare_deploy = pipeline(*ray_launcher) | |||
| with lazyllm.pipeline() as ppl: | |||
There was a problem hiding this comment.
这个情况可以试一下post_action
parall_launcher = [lazyllm.pipeline(sleep_moment, launcher) for launcher in ray_launcher[1:]]
self._prepare_deploy = lazyllm.pipeline(ray_launcher[0], post_action=(lazyllm.parallel(*parall_launcher) if len(parall_launcher) else None))
lwj-st
added a commit
to LazyAGI/LazyLLM-Env
that referenced
this pull request
Mar 6, 2025
lwj-st
added a commit
to LazyAGI/LazyLLM-Env
that referenced
this pull request
Mar 12, 2025
lwj-st
added a commit
to LazyAGI/LazyLLM-Env
that referenced
this pull request
Mar 12, 2025
# for free
to join this conversation on GitHub.
Already have an account?
# to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.