mirror of
https://github.com/PaddlePaddle/FastDeploy.git
synced 2025-12-24 13:28:13 +08:00
11 lines
235 B
YAML
11 lines
235 B
YAML
enable_mm: True
|
|
max_model_len: 32768
|
|
max_num_seqs: 128
|
|
gpu_memory_utilization: 0.9
|
|
kv_cache_ratio: 0.71
|
|
tensor_parallel_size: 1
|
|
enable_chunked_prefill: True
|
|
max_num_batched_tokens: 384
|
|
quantization: wint4
|
|
reasoning_parser: ernie-45-vl
|