Download evaluate_vllm.sh from chibifire/EditScore-code: direct link, hf CLI and curl.
- Browser
- Download file 523 Bytes
-
https://huggingface.co/chibifire/EditScore-code/resolve/main/evaluate_vllm.sh
- Command line
-
hf download hf://chibifire/EditScore-code/evaluate_vllm.sh
-
curl -L -o evaluate_vllm.sh https://huggingface.co/chibifire/EditScore-code/resolve/main/evaluate_vllm.sh
523 Bytes
| # !/bin/bash | |
| SHELL_FOLDER=$(cd "$(dirname "$0")";pwd) | |
| cd $SHELL_FOLDER | |
| python evaluation.py \ | |
| --benchmark_dir EditScore/EditReward-Bench \ | |
| --result_dir results/EditScore-7B \ | |
| --backbone qwen25vl_vllm \ | |
| --model_name_or_path Qwen/Qwen2.5-VL-7B-Instruct \ | |
| --lora_path EditScore/EditScore-7B \ | |
| --score_range 25 \ | |
| --max_workers 1 \ | |
| --max_model_len 4096 \ | |
| --max_num_seqs 1 \ | |
| --max_num_batched_tokens 4096 \ | |
| --tensor_parallel_size 1 \ | |
| --num_pass 1 | |
| python calculate_statistics.py \ | |
| --result_dir results/EditScore-7B/qwen25vl_vllm |