apple

/

FastVLM-7B

like 270

Follow

Apple 6.7k

• Add input_ids parsing into the generate function, commonly used in 🤗 pipelines

Example:

# Load models
tok = AutoTokenizer.from_pretrained("apple/FastVLM-7B", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    "apple/FastVLM-7B",
    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
    device_map="auto",
    trust_remote_code=True,
)

# Build prompt
messages = [
    {"role": "user", "content": "Describe San Francisco"}
]

# Tokenize
inputs = tokenizer.apply_chat_template(
      messages,
      add_generation_prompt=True,
      tokenize=True,
      return_dict=True,
      return_tensors="pt",
 ).to(model.device)

# Generate  
with torch.no_grad():
    out = model.generate(
        **inputs,
        max_new_tokens=128,
    )

print(tokenizer.decode(out[0], skip_special_tokens=True))