EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Adopt inductor fusion and define quantization fusion pass (#4168)

Browse Source 
### What this PR does / why we need it?
The main goal of this PR to alleviate the high maintenance burden from
model duplication when we are going to do the model optimization. Some
of our optimized models diverges a little from the vllm's modeling, but
needs to rewrite several part of original one, brings negligible
maintenance bruden to the vllm-ascend.In order to solve that, we propose
to leverage `torch.compile` and `inductor pattern matcher`,
automatically fuse the pattern we want to merge. For more details can
refer to the RFC https://github.com/vllm-project/vllm-ascend/issues/4239

This pr integrates `AddRMSNorm` and the `Quant` operator, which can
improve the inference speed of models using `w8a8 `quantization.

### Does this PR introduce _any_ user-facing change?
Yes, add new additional_config

### How was this patch tested?
```python
def main():
    prompts = [
        "The president of the United States is Mr.",
    ]

    # Create a sampling params object.
    sampling_params = SamplingParams(max_tokens=100, temperature=0.6, top_k=40, top_p=0.95)
    # Create an LLM.
    llm = LLM(
        model="/root/.cache/modelscope/hub/models/vllm-ascend/Qwen3-8B-W8A8",
              # enforce_eager=True,
              tensor_parallel_size=1,
              trust_remote_code=True,
              gpu_memory_utilization=0.7,
              quantization="ascend",
              )

    # Generate texts from the prompts.
    outputs = llm.generate(prompts, sampling_params)
    for output in outputs:
        prompt = output.prompt
        generated_text = output.outputs[0].text
        print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
```

```text
Prompt: 'The president of the United States is Mr.', Generated text: ' Trump. The president of the United States is Mr. Biden. Which of the following statements is correct? \n\nA. Mr. Trump is Mr. Biden.  \nB. Mr. Trump is not Mr. Biden.  \nC. The president of the United States is not Mr. Trump.  \nD. The president of the United States is not Mr. Biden.\n\nThe question presents a contradiction: it states that "The president of the United States is Mr. Trump" and "The president of'
```


- vLLM version: 86e178f7c4d8c3b0eaf3c8e3f810a83f63b90e24
- vLLM main:
86e178f7c4

---------

Signed-off-by: Icey <1790571317@qq.com>
Signed-off-by: wxsIcey <1790571317@qq.com>
This commit is contained in:
Icey
2025-12-04 10:29:48 +08:00
committed by GitHub
parent c4a71fc6d5
commit 178ca1607e
13 changed files with 593 additions and 267 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
vllm_ascend/ascend_forward_context.py
View File

@@ -159,25 +159,6 @@ def set_ascend_forward_context(
forward_context.weight_prefetch_method = weight_prefetch_method
forward_context.is_mtp_model = is_mtp_model
# TODO(rjg-lyh): The current implementation is somewhat brute force and not elegant.
# It will be improved later by implementing operator fusion through the FX graph.
#
# set for addrmsnorm+quant fusion.
# this optim now just support dense models due to the specific operators used.
# Once the necessary conditions are met, support for MOE models will also be added.
from vllm_ascend.quantization.quant_config import AscendQuantConfig
model_type_scope = ["llama", "qwen2", "qwen3", "qwen3_moe"]
addrmsnorm_quant_fusion_enabled = isinstance(vllm_config.quant_config, AscendQuantConfig) and \
vllm_config.model_config.hf_config.model_type in model_type_scope and \
forward_context.layer_idx is not None
if addrmsnorm_quant_fusion_enabled:
forward_context.model_instance = model_instance
forward_context.num_hidden_layers = vllm_config.model_config.hf_config.num_hidden_layers
forward_context.fusion_linear = "gate_up_dense" if forward_context.layer_idx == 0 else "qkv_dense"
if vllm_config.model_config.hf_config.model_type == "qwen3_moe":
forward_context.fusion_linear = "gate_moe" if forward_context.layer_idx == 0 else "qkv_moe"
forward_context.addrmsnorm_quant_fusion_enabled = addrmsnorm_quant_fusion_enabled
if num_tokens is None and attn_metadata is not None:
num_tokens = attn_metadata.num_actual_tokens