Merge pull request #14 from meta-llama/lmm_infer

Create multi_modal_infer.py

recipes/quickstart/inference/local_inference/README.md

@@ -1,5 +1,12 @@
 # Local Inference
 
+For Multi-Modal inference we have added [multi_modal_infer.py](multi_modal_infer.py) which uses the transformers library
+
+The way to run this would be
+```
+python multi_modal_infer.py --image_path "./resources/image.jpg" --prompt_text "Describe this image" --temperature 0.5 --top_p 0.8 --model_name "meta-llama/Llama-3.2-11B-Vision-Instruct"
+```
+
 For local inference we have provided an [inference script](inference.py). Depending on the type of finetuning performed during training the [inference script](inference.py) takes different arguments.
 To finetune all model parameters the output dir of the training has to be given as --model_name argument.
 In the case of a parameter efficient method like lora the base model has to be given as --model_name and the output dir of the training has to be given as --peft_model argument.
@@ -87,4 +94,4 @@ python inference.py --model_name <training_config.output_dir> --prompt_file <tes
 
 ## Inference on large models like Meta Llama 405B
 The FP8 quantized variants of Meta Llama (i.e. meta-llama/Meta-Llama-3.1-405B-FP8 and meta-llama/Meta-Llama-3.1-405B-Instruct-FP8) can be executed on a single node with 8x80GB H100 using the scripts located in this folder.
-To run the unquantized Meta Llama 405B variants (i.e. meta-llama/Meta-Llama-3.1-405B and meta-llama/Meta-Llama-3.1-405B-Instruct) we need to use a multi-node setup for inference. The llama-recipes inference script currently does not allow multi-node inference. To run this model you can use vLLM with pipeline and tensor parallelism as showed in [this example](../../../3p_integrations/vllm/README.md).
+To run the unquantized Meta Llama 405B variants (i.e. meta-llama/Meta-Llama-3.1-405B and meta-llama/Meta-Llama-3.1-405B-Instruct) we need to use a multi-node setup for inference. The llama-recipes inference script currently does not allow multi-node inference. To run this model you can use vLLM with pipeline and tensor parallelism as showed in [this example](../../../3p_integrations/vllm/README.md).

recipes/quickstart/inference/local_inference/multi_modal_infer.py

@@ -0,0 +1,66 @@
+import os
+import sys
+import argparse
+from PIL import Image as PIL_Image
+import torch
+from transformers import MllamaForConditionalGeneration, MllamaProcessor
+
+
+# Constants
+DEFAULT_MODEL = "meta-llama/Llama-3.2-11B-Vision-Instruct"
+
+
+def load_model_and_processor(model_name: str, hf_token: str):
+    """
+    Load the model and processor based on the 11B or 90B model.
+    """
+    model = MllamaForConditionalGeneration.from_pretrained(model_name, device_map="auto", torch_dtype=torch.bfloat16, token=hf_token)
+    processor = MllamaProcessor.from_pretrained(model_name, token=hf_token)
+    return model, processor
+
+
+def process_image(image_path: str) -> PIL_Image.Image:
+    """
+    Open and convert an image from the specified path.
+    """
+    if not os.path.exists(image_path):
+        print(f"The image file '{image_path}' does not exist.")
+        sys.exit(1)
+    with open(image_path, "rb") as f:
+        return PIL_Image.open(f).convert("RGB")
+
+
+def generate_text_from_image(model, processor, image, prompt_text: str, temperature: float, top_p: float):
+    """
+    Generate text from an image using the model and processor.
+    """
+    conversation = [
+        {"role": "user", "content": [{"type": "image"}, {"type": "text", "text": prompt_text}]}
+    ]
+    prompt = processor.apply_chat_template(conversation, add_generation_prompt=True, tokenize=False)
+    inputs = processor(prompt, image, return_tensors="pt").to(model.device)
+    output = model.generate(**inputs, temperature=temperature, top_p=top_p, max_new_tokens=512)
+    return processor.decode(output[0])[len(prompt):]
+
+
+def main(image_path: str, prompt_text: str, temperature: float, top_p: float, model_name: str, hf_token: str):
+    """
+    Call all the functions. 
+    """
+    model, processor = load_model_and_processor(model_name, hf_token)
+    image = process_image(image_path)
+    result = generate_text_from_image(model, processor, image, prompt_text, temperature, top_p)
+    print("Generated Text: " + result)
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description="Generate text from an image and prompt using the 3.2 MM Llama model.")
+    parser.add_argument("--image_path", type=str, help="Path to the image file")
+    parser.add_argument("--prompt_text", type=str, help="Prompt text to describe the image")
+    parser.add_argument("--temperature", type=float, default=0.7, help="Temperature for generation (default: 0.7)")
+    parser.add_argument("--top_p", type=float, default=0.9, help="Top p for generation (default: 0.9)")
+    parser.add_argument("--model_name", type=str, default=DEFAULT_MODEL, help=f"Model name (default: '{DEFAULT_MODEL}')")
+    parser.add_argument("--hf_token", type=str, required=True, help="Hugging Face token for authentication")
+
+    args = parser.parse_args()
+    main(args.image_path, args.prompt_text, args.temperature, args.top_p, args.model_name, args.hf_token)