[Misc] Throw error on unintended access to scheduler_config.max_model_len

[frank-wei]

Purpose

There is a risk when people unintentionally try to access scheduler_config.max_model_len and it is returned as 8192. SInce it is removed in #28733, we should give people AttributeError as alert.

Test Plan

Test Result

Before the fix:

from vllm.config import VllmConfig
vllm_config = VllmConfig()
INFO 11-30 22:40:59 [scheduler.py:207] Chunked prefill is enabled with max_num_batched_tokens=2048.
vllm_config.scheduler_config.max_model_len
8192

After the fix:

from vllm.config import VllmConfig
vllm_config = VllmConfig()
INFO 11-30 22:40:08 [scheduler.py:207] Chunked prefill is enabled with max_num_batched_tokens=2048.
vllm_config.scheduler_config.max_model_len
Traceback (most recent call last):
File "", line 1, in
File "/data/users/wwei6/gitrepos/vllm-main-frank-branch/vllm/config/scheduler.py", line 280, in getattribute
raise AttributeError(
AttributeError: max_model_len is an init-only parameter.

---

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan, such as providing test command.
• The test results, such as pasting the results comparison before and after, or e2e results
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.
• (Optional) Release notes update. If your change is user facing, please update the release notes draft in the Google Doc.

[chatgpt-codex-connector]

Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits.

[gemini-code-assist]

Code Review

The pull request correctly addresses the issue of unintended access to InitVar attributes in SchedulerConfig by implementing a __getattribute__ method. This prevents incorrect values from being returned and guides users to the correct ModelConfig for these parameters. This is a good defensive programming practice.

[DarkLight1337]

@hmellor I thought InitVar would do that already...

[frank-wei]

@hmellor I thought InitVar would do that already...

@DarkLight1337 , I thought too but seems that @config decoration will leverage Pydantic model and it stores it as attribute indeed.