graphpatch 0.1.0#

Overview#

graphpatch is a library for activation patching on PyTorch neural network models. You use it by first wrapping your model in a PatchableGraph and then running operations in a context created by PatchableGraph.patch():

model = GPT2LMHeadModel.from_pretrained(
   "gpt2-xl",
   device_map="auto",
   load_in_8bit=True,
   torch_dtype=torch.float16
)
tokenizer = AutoTokenizer.from_pretrained("gpt2-xl")
inputs = tokenizer(
   "The Eiffel Tower, located in", return_tensors="pt", padding=False
).to(torch.device("cuda"))
# Note that all arguments after the first are forwarded as example inputs
# to the model during compilation; use_cache and return_dict are arguments
# to GPT2LMHeadModel, not graphpatch-specific.
pg = PatchableGraph(model, **inputs, use_cache=False, return_dict=False)
# Applies two patches to the multiplication result within the activation function
# of the MLP in the 18th transformer layer. ProbePatch records the last observed value
# at the given node, while ZeroPatch zeroes out the value seen by downstream computations.
with pg.patch("transformer.h_17.mlp.act.mul_3": [probe := ProbePatch(), ZeroPatch()]):
   output = pg(**inputs)
# Patches are applied in order. probe.activation holds the value prior
# to ZeroPatch zeroing it out.
print(probe.activation)

In contrast to other approaches, graphpatch can patch (or record) any intermediate Tensor value without manual modification of the underlying model’s code. This flexibility and automaticity come with a tradeoff: in some cases, it may be difficult to find the precise location of a desired intervention. See Working with graphpatch for some tips on navigating the generated graphs.

Note that graphpatch activation patches are compatible with AutoGrad! This means that, for example, you can perform optimizations over the value parameter to AddPatch:

delta = torch.zeros(size, requires_grad=True, device="cuda")
optimizer = torch.optim.Adam([delta], lr=0.5)
for _ in range(num_steps):
   with graph.patch({node_name: AddPatch(value=delta)):
      logits = graph(**prompt_inputs)
   loss = my_loss_function(logits)
   loss.backward()
   optimizer.step()

For a practical usage example, see the demo of using graphpatch to replicate ROME.

Prerequisites#

The only mandatory requirement is torch>=2. Version 2+ is required because graphpatch leverages torch.compile(), which was introduced in 2.0.0, to extract computational graphs from models. CUDA support is not required.

Python 3.8–3.11 are supported. Note that torch versions prior to 2.1.0 do not support compilation on Python 3.11; you will get an exception when trying to use graphpatch with such a configuration. No version of torch yet supports compilation on Python 3.12.

graphpatch is theoretically compatible with any model in Huggingface’s transformers library, but note that there may be edge cases in specific model code that it can’t yet handle. graphpatch is tested against and known to work with the transformers implementations of Llama and GPT2.

graphpatch is compatible with models loaded via accelerate and with 8-bit parameters quantized by bitsandbytes. This means that you can run graphpatch on multiple GPU’s and/or with quantized inference very easily on models provided by transformers:

model = LlamaForCausalLM.from_pretrained(
   model_path, device_map="auto", load_in_8bit=True, torch_dtype=torch.float16
)
pg = PatchableGraph(model, **example_inputs)

Installation#

graphpatch is available on PyPI, and can be installed via pip:

pip install graphpatch

Note that you will likely want to do this in an environment that already has torch, since pip may not resolve torch to a CUDA-enabled version by default. You don’t need to do anything special to make graphpatch compatible with transformers, accelerate, and bitsandbytes; their presence is detected at run-time. However, for convenience, you can install graphpatch with the “transformers” extra, which will install known compatible versions of these libraries along with some of their optional dependencies that are otherwise mildly inconvenient to set up:

pip install graphpatch[transformers]

Alternatives#

Module hooks are built in to torch and can be used for activation patching. You can even add them to existing models without modifying their code. However, this will only give you access to module inputs and outputs; accessing or patching intermediate values still requires a manual rewrite.

TransformerLens provides the HookPoint class, which can record and patch intermediate activations. However, this requires manually rewriting your model’s code to wrap the values you want to make patchable.

TorchLens records and outputs visualizations for every intermediate activation. However, it is currently unable to perform any activation patching.

Documentation index#

Full index