| 12345678910111213141516171819202122 |
- from transformers import AutoConfig
- from gpustack.utils.hub import get_hf_text_config, get_max_model_len
- def test_get_max_model_len():
- hf_model_lengths = {
- "Qwen/Qwen2-0.5B-Instruct": 32768,
- "Qwen/Qwen2-VL-7B-Instruct": 32768,
- "tiiuae/falcon-7b": 2048,
- "microsoft/Phi-3.5-mini-instruct": 131072,
- "llava-hf/llava-v1.6-mistral-7b-hf": 32768,
- "unsloth/Llama-3.2-11B-Vision-Instruct": 131072,
- "THUDM/glm-4-9b-chat-1m": 1048576,
- }
- for model_name, expected_max_model_len in hf_model_lengths.items():
- config = AutoConfig.from_pretrained(model_name, trust_remote_code=True)
- pretrained_or_hf_text_config = get_hf_text_config(config)
- assert (
- get_max_model_len(pretrained_or_hf_text_config) == expected_max_model_len
- ), f"max_model_len mismatch for {model_name}"
|
