vllm
vllm copied to clipboard
[Bug]: flash_attn_varlen_func() got an unexpected keyword argument 'softcap'
Your current environment
I built my flash-attn from the source. The version is 2.6.3
🐛 Describe the bug
When runing following code:
self.policy_model = LLM(
model="OpenGVLab/InternVL2-8B",
trust_remote_code=True,
max_num_seqs=5,
gpu_memory_utilization=0.40,
)
Raise the following error:
[rank0]: TypeError: flash_attn_varlen_func() got an unexpected keyword argument 'softcap'