Safetensors
matveymih commited on
Commit
e109ff2
1 Parent(s): 0c459fe

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +7 -7
README.md CHANGED
@@ -91,13 +91,13 @@ from models import CLIPVisionTower
91
  DEVICE = "cuda:0"
92
  PROMPT = "This is a dialog with AI assistant.\n"
93
 
94
- tokenizer = AutoTokenizer.from_pretrained("AIRI-Institute/OmniFusion", subfolder="OmniMistral-tokenizer", use_fast=False)
95
- model = AutoModelForCausalLM.from_pretrained("AIRI-Institute/OmniFusion", subfolder="OmniMistral-model", torch_dtype=torch.bfloat16, device_map=DEVICE)
96
 
97
- hf_hub_download(repo_id="AIRI-Institute/OmniFusion", filename="projection", local_dir='./')
98
- hf_hub_download(repo_id="AIRI-Institute/OmniFusion", filename="special_embeddings.pt", local_dir='./')
99
- projection = torch.load("projection", map_location=DEVICE)
100
- special_embs = torch.load("special_embeddings.pt", map_location=DEVICE)
101
 
102
  clip = CLIPVisionTower("openai/clip-vit-large-patch14-336")
103
  clip.load_model()
@@ -149,7 +149,7 @@ def gen_answer(model, tokenizer, clip, projection, query, special_embs, image=No
149
  return generated_texts
150
 
151
  img_url = "https://i.pinimg.com/originals/32/c7/81/32c78115cb47fd4825e6907a83b7afff.jpg"
152
- question = "who is the author?"
153
  img = Image.open(urlopen(img_url))
154
 
155
  answer = gen_answer(
 
91
  DEVICE = "cuda:0"
92
  PROMPT = "This is a dialog with AI assistant.\n"
93
 
94
+ tokenizer = AutoTokenizer.from_pretrained("AIRI-Institute/OmniFusion", subfolder="OmniMistral-v1_1/tokenizer", use_fast=False)
95
+ model = AutoModelForCausalLM.from_pretrained("AIRI-Institute/OmniFusion", subfolder="OmniMistral-v1_1/tuned-model", torch_dtype=torch.bfloat16, device_map=DEVICE)
96
 
97
+ hf_hub_download(repo_id="AIRI-Institute/OmniFusion", filename="OmniMistral-v1_1/projection.pt", local_dir='./')
98
+ hf_hub_download(repo_id="AIRI-Institute/OmniFusion", filename="OmniMistral-v1_1/special_embeddings.pt", local_dir='./')
99
+ projection = torch.load("OmniMistral-v1_1/projection.pt", map_location=DEVICE)
100
+ special_embs = torch.load("OmniMistral-v1_1/special_embeddings.pt", map_location=DEVICE)
101
 
102
  clip = CLIPVisionTower("openai/clip-vit-large-patch14-336")
103
  clip.load_model()
 
149
  return generated_texts
150
 
151
  img_url = "https://i.pinimg.com/originals/32/c7/81/32c78115cb47fd4825e6907a83b7afff.jpg"
152
+ question = "What is the sky color on this image?"
153
  img = Image.open(urlopen(img_url))
154
 
155
  answer = gen_answer(