Skip to content

v0.5.0: Tutorial, Better contrastive attribution, 4-bit/Petals support and more
Compare
Choose a tag to compare
@github-actions github-actions released this 07 Dec 09:49
· 59 commits to refs/heads/main since this release
v0.5.0
110e038
This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
GPG key ID: 4AEE18F83AFDEB23
Expired
Learn about vigilant mode.

📄 New Tutorial and Better Documentation

  • A new quickstart tutorial is available in the repository, introducing feature attribution methods and showcasing basic and more advanced Inseq use-cases.

  • Documentation now uses the Sphinx furo theme

  • A new utility function inseq.explain was introduced to visualised docs associated with string identifiers used for attribution methods, step functions and aggregators

import inseq

inseq.explain("saliency")
>>> Saliency attribution method.

    Reference implementation:
    `https://captum.ai/api/saliency.html <https://captum.ai/api/saliency.html>`__.

🔀 More Flexible and Intuitive Contrastive Attribution (#193, #195, #207, #228)

  • Contrastive attribution functions now support original and contrastive targets of different lengths, using right-side alignment of tokens by default to simplify usage for studies using preceding context as contrastive option.

  • Contrastive source and target inputs can be specified as string inputs to model.attribute when using a contrastive step function of attribution target using the contrast_sources and contrast_targets arguments (see docs)

  • Custom alignments can be provided for contrastive step functions to compare specific step pairs using the contrast_targets_alignments argument in model.attribute. Using ”auto” uses a multilingual LaBSE encoder for creating alignments using the AWESOME approach (useful for generation tasks preserving semantic equivalence, e.g. machine translation)

  • The is_attributed_fn argument in StepFunctionBaseArgs can be used to customize the behavior of step functions in the attributed or the regular cases.

Refer to the quickstart tutorial for examples of contrastive attribution.

🤗 Support for Distributed and 4-bit Models (#186, #205)

Towards the goal of democratizing the access to interpretability methods for analyzing state-of-the-art models, Inseq now supports attribution of distributed language models from the Petals library and 4-bit quantized LMs from the transformersbitsandbytes integration using load_in_4bit=True, with the added flexibility of the Inseq API.

Example of contrastive gradient attribution of a distributed LLaMA 65B model:

from petals import AutoDistributedModelForCausalLM

model_name = "enoch/llama-65b-hf"
model = AutoDistributedModelForCausalLM.from_pretrained(model_name).cuda()
inseq_model = inseq.load_model(model, "saliency")
prompt = (
    "Option 1: Take a 50 minute bus, then a half hour train, and finally a 10 minute bike ride.\n"
    "Option 2: Take a 10 minute bus, then an hour train, and finally a 30 minute bike ride.\n"
    "Which of the options above is faster to get to work?\n"
    "Answer: Option”
)
out = inseq_model.attribute(
    prompt,
    prompt + "1",
    attributed_fn="contrast_prob_diff",
    contrast_targets=prompt + "2",
)

Refer to the doc guide for more details.

🔍 New Step Functions and Attribution Methods (#182, #222, #223)

The following step functions were added as pre-registered in this release:

  • logits: Logits of the target token.
  • contrast_logits/contrast_prob: Logits/probabilities of the target token when different contrastive inputs are provided to the model. Equivalent to logits/probability when no contrastive inputs are provided.
  • pcxmi: Point-wise Contextual Cross-Mutual Information (P-CXMI) for the target token given original and contrastive contexts (Yin et al. 2021).
  • kl_divergence: KL divergence of the predictive distribution given original and contrastive contexts. Can be restricted to most likely target token options using the top_k and top_p parameters.
  • in_context_pvi: In-context Pointwise V-usable Information (PVI) to measure the amount of contextual information used in model predictions (Lu et al. 2023).
  • top_p_size: The number of tokens with cumulative probability greater than top_p in the predictive distribution of the model.

The following attribution method was also added:

💥 Breaking Changes

  • The contrast_ids and contrast_attention_mask parameters in model.attribute for contrastive step functions and attribution targets are deprecated in favor of contrast_sources and contrast_targets.

  • Extraction and aggregation of attention weights from the attention method is now handled post-hoc via Aggregator classes, making it uniform to the API adopted for other attribution methods.

All Merged PRs

🚀 Features

  • Attributed behavior for contrastive step functions (#228) @gsarti
  • Step functions fixes, add in_context_pvi (#223) @gsarti
  • Add Sequential IG method (#222) @gsarti
  • Allow contrastive attribution with shorter contrastive targets (#207) @gsarti
  • Add top_p_size step fn, StepFunctionArgs class (#206) @gsarti
  • Support petals distributed model classes (#205) @gsarti
  • Custom alignment of contrast_targets for contrastive attribution methods (#195) @gsarti
  • Tokens diff view for contrastive attribution methods (#193) @gsarti
  • Handle .to for 4bit quantized models (#186) @g8a9
  • Aggregation functions, named aggregators, contrastive context step functions, inseq.explain (#182) @gsarti
  • Target prefix-constrained generation (#172) @gsarti

🔧 Fixes & Refactoring

📝 Documentation and Tutorials

👥 List of contributors

@gsarti, @DanielSc4, @carschno, @g8a9 and @lsickert

Contributors

carschno, gsarti, and 3 other contributors
Assets 3
Loading