Add support for conversation-template argument for openai endpoint #1294
Conversation
This file contains hidden or 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
…o a json template. Added to quickstart.rst and example json file in the examples folder.
This was referenced Oct 9, 2023
|
Going to cancel this PR, and work on another which properly implements the HF chat templates within the tokenizer. |
amy-why-3459
pushed a commit
to amy-why-3459/vllm
that referenced
this pull request
Sep 15, 2025
### What this PR does / why we need it?
1. Add MTP dummy_run, and adapt main model dummy_run when mtp is enabled
2. Adapt main model torchair graph mode, when mtp is enabled
3. mtp model torchair graph mode will be supported in the future
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
This patch is tested by
`vllm-ascend/tests/long_term/spec_decode_v1/test_v1_mtp_correctness.py`
### Usage
online example
```shell
export VLLM_USE_V1=1
export VLLM_ENABLE_MC2=1
export VLLM_VERSION=0.9.1
export ASCEND_LAUNCH_BLOCKING=0
python -m vllm.entrypoints.openai.api_server \
--model="/model_weight_path" \
--trust-remote-code \
--max-model-len 40000 \
--tensor-parallel-size 4 \
--data_parallel_size 4 \
--enable_expert_parallel \
--served-model-name deepseekr1 \
--quantization ascend \
--host 0.0.0.0 \
--port 1234 \
--speculative-config '{"num_speculative_tokens": 1, "method":"deepseek_mtp"}' \
--additional-config '{"ascend_scheduler_config":{"enabled":true,"enable_chunked_prefill":false},"torchair_graph_config":{"enabled":true}}' \
--gpu_memory_utilization 0.95
```
offline example
```
lm = LLM(
model="/model_weight_path",
tensor_parallel_size=16,
max_num_seqs=128,
gpu_memory_utilization=0.95,
distributed_executor_backend="mp",
enable_expert_parallel=True,
speculative_config={
"method": "deepseek_mtp",
"num_speculative_tokens": 1,
},
trust_remote_code=True,
enforce_eager=False,
additional_config = {
'torchair_graph_config': {
'enabled': True,
'enable_multistream_shared_expert': False
},
"ascend_scheduler_config": {
"enabled": True
},
}
)
```
Signed-off-by: xuyexiong <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in 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.
Here is a PR for support for conversation-templates setup as json files which can be specified for a model upon starting the api. Just create your template, and pass in the path with the --conversation-template my_template.json argument.
There was no where for me to add support for this in the vLLM api, so I only added it to the OpenAI api.
Added an example, and updated the quickstart part of the readme (the only place that talked about the other arguments)