Files
FastDeploy/benchmarks/yaml/eb45-vl-32k-wint4-h800-tp8.yaml
2025-06-29 23:29:37 +00:00

12 lines
285 B
YAML

enable_mm: True
max_model_len: 32768
max_num_seqs: 56
gpu_memory_utilization: 0.8
kv_cache_ratio: 0.8
tensor_parallel_size: 8
quantization: wint4
limit_mm_per_prompt: '{"image": 100, "video": 100}'
enable_chunked_prefill: True
max_num_batched_tokens: 384
reasoning_parser: ernie-45-vl