add VLM: monkey

vlmeval/config.py

@@ -43,6 +43,7 @@
  'cogvlm-grounding-generalist':partial(CogVlm, name='cogvlm-grounding-generalist',tokenizer_name ='lmsys/vicuna-7b-v1.5'),
  'cogvlm-chat':partial(CogVlm, name='cogvlm-chat',tokenizer_name ='lmsys/vicuna-7b-v1.5'),
  'sharedcaptioner':partial(SharedCaptioner, model_path='Lin-Chen/ShareCaptioner'),
+ 'monkey':partial(Monkey, model_path='echo840/Monkey'),
 }
 
 api_models = {

vlmeval/vlm/__init__.py

@@ -15,3 +15,4 @@
 from .llava_xtuner import LLaVA_XTuner
 from .cogvlm import CogVlm
 from .sharedcaptioner import SharedCaptioner
+from .monkey import Monkey

vlmeval/vlm/monkey.py

@@ -0,0 +1,43 @@
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import warnings
+import os.path as osp
+from vlmeval.smp import isimg
+import re
+
+class Monkey:
+
+ INSTALL_REQ = False
+
+ def __init__(self, model_path='echo840/Monkey', **kwargs):
+ assert model_path is not None
+ self.model_path = model_path
+ self.tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
+ self.model = AutoModelForCausalLM.from_pretrained(model_path, device_map='cuda', trust_remote_code=True).eval()
+ self.kwargs = kwargs
+ warnings.warn(f"Following kwargs received: {self.kwargs}, will use as generation config. ")
+ torch.cuda.empty_cache()
+
+ def generate(self, image_path, prompt, dataset=None):
+ cur_prompt = f'<img>{image_path}</img> {prompt} Answer: '
+ input_ids = self.tokenizer(cur_prompt, return_tensors='pt', padding='longest')
+ attention_mask = input_ids.attention_mask
+ input_ids = input_ids.input_ids
+
+ output_ids = self.model.generate(
+ input_ids=input_ids.cuda(),
+ attention_mask=attention_mask.cuda(),
+ do_sample=False,
+ num_beams=1,
+ max_new_tokens=512,
+ min_new_tokens=1,
+ length_penalty=3,
+ num_return_sequences=1,
+ output_hidden_states=True,
+ use_cache=True,
+ pad_token_id=self.tokenizer.eod_id,
+ eos_token_id=self.tokenizer.eod_id,
+ )
+ response = self.tokenizer.decode(output_ids[0][input_ids.size(1):].cpu(), skip_special_tokens=True).strip()
+
+ return response