Experimenting With Transformers & Nokens

Mon, 04 Mar 2024 00:00:00 +0000

Sometimes a single word isn’t enough. What if you could blend tokens or use non-tokens (aka nokens¹)?

Let’s explore these ideas, learn more about transformers: the architecture and Hugging Face library.

Installing & Setup

I’m going to use Microsoft’s Phi-2 as it is smaller than Mistral / Llama2. The ideas should transfer to other transformer models.

After pip install transformers torch, we can load our language model to experiment with using the Hugging Face library.

Rorschach Test For LLaVA LLM

Sat, 30 Dec 2023 00:00:00 +0000

LLaVA works by teaching an LLM (Llama2) to “understand” embeddings from an image model (CLIP). What happens if you replace the embeddings with randomness and ask the LLM questions about what it sees?

Hacking LLaVA

After loading the LLaVA model, we can inject a function that runs after clip embed but before projection to the LLMs token latent space.

# load LLaVA using HF transformers
tokenizer, model, image_processor, context_len = load_pretrained_model(
 "liuhaotian/llava-v1.5-13b",
 model_name="llava-v1.5-13b",
 model_base=None,
 load_8bit=False,
 load_4bit=False)

# inject hook that allows modifying the generated clip embed
tower = model.get_vision_tower() 
tower.register_forward_hook(random_forward_hook)

# if global alter_seed set, override embed with randn of same shape
alter_seed = None
def random_forward_hook(module, input, output):
 if not alter_seed is None:
 print("overriding image embedding with seed: ", alter_seed)
 torch.manual_seed(alter_seed)
 output[0] = torch.randn_like(output[0])
 return output

You can try it out by running llava-lies on replicate. This model still asks for an image, to simplify the modification of LLaVA’s code. The hook random_forward_hook above replaces the output with normalized seeded noise.

Llm on anotherjesse.com

Experimenting With Transformers & Nokens

Installing & Setup

Rorschach Test For LLaVA LLM

Hacking LLaVA