[Doc] Upgrade multi-node doc (#4365)

### What this PR does / why we need it?
When we are using `Ascend scheduler`, the param `max_num_batched_tokens`
should be larger than `max_model_len`, otherwise, will encountered the
follow error:
```shell
Value error, Ascend scheduler is enabled without chunked prefill feature. Argument max_num_batched_tokens (4096) is smaller than max_model_len (32768). This effectively limits the maximum sequence length to max_num_batched_tokens and makes vLLM reject longer sequences. Please increase max_num_batched_tokens or decrease max_model_len. [type=value_error, input_value=ArgsKwargs((), {'model_co...g': {'enabled': True}}}), input_type=ArgsKwargs]
```

### Does this PR introduce _any_ user-facing change?
Users/Developers who running the model according to the
[tutorial](https://docs.vllm.ai/projects/ascend/en/latest/tutorials/multi_node.html),
the parameters can be specified correctly.

### How was this patch tested?

- vLLM version: v0.11.0
- vLLM main:
2918c1b49c

---------

Signed-off-by: wangli <wangli858794774@gmail.com>
This commit is contained in:
Li Wang
2025-11-24 10:57:50 +08:00
committed by GitHub
parent b34f195cc8
commit b5f7a83927
3 changed files with 9 additions and 9 deletions

View File

@@ -131,9 +131,9 @@ vllm serve vllm-ascend/DeepSeek-V3.1-W8A8 \
--served-model-name deepseek_v3.1 \
--enable-expert-parallel \
--max-num-seqs 16 \
--max-model-len 32768 \
--max-model-len 8192 \
--quantization ascend \
--max-num-batched-tokens 4096 \
--max-num-batched-tokens 8192 \
--trust-remote-code \
--no-enable-prefix-caching \
--gpu-memory-utilization 0.9 \
@@ -176,8 +176,8 @@ vllm serve vllm-ascend/DeepSeek-V3.1-W8A8 \
--quantization ascend \
--served-model-name deepseek_v3.1 \
--max-num-seqs 16 \
--max-model-len 32768 \
--max-num-batched-tokens 4096 \
--max-model-len 8192 \
--max-num-batched-tokens 8192 \
--enable-expert-parallel \
--trust-remote-code \
--no-enable-prefix-caching \

View File

@@ -88,8 +88,8 @@ vllm serve /home/cache/weights/Kimi-K2-Instruct-W8A8 \
--tensor-parallel-size 8 \
--enable-expert-parallel \
--max-num-seqs 16 \
--max-model-len 32768 \
--max-num-batched-tokens 4096 \
--max-model-len 8192 \
--max-num-batched-tokens 8192 \
--trust-remote-code \
--no-enable-prefix-caching \
--gpu-memory-utilization 0.9 \
@@ -130,9 +130,9 @@ vllm serve /home/cache/weights/Kimi-K2-Instruct-W8A8 \
--tensor-parallel-size 8 \
--served-model-name kimi \
--max-num-seqs 16 \
--max-model-len 32768 \
--max-model-len 8192 \
--quantization ascend \
--max-num-batched-tokens 4096 \
--max-num-batched-tokens 8192 \
--enable-expert-parallel \
--trust-remote-code \
--no-enable-prefix-caching \