likenneth/honest_llama

• highreadme#1

  Add a concise, problem-solution opening statement to the README

  Why:

  CURRENT

  ### Update 08/24/2024
  With the release of LLaMA-3 models, I decided to replicate ITI on a suite of LLaMA models for easy comparison. I've recorded the results in `iti_replication_results.md` and uploaded the ITI baked-in models to HuggingFace here. Note that the ITI baked-in models and ITI applied to base models is not exactly a one-to-one comparison due to slight differences in when the activations are edited. The ITI baked-in models have the activation differences hardcoded into their attention biases. For more precise editing, consider only using the models' attention biases when processing tokens after the input prompt, to be more faithful to the original ITI method.
  
  -- Justin Ji @jujipotle
  
  ### Update 01/26/2024 :fire::fire:
  
  Zen provided this really cool library called pyvene that can be used to load Inference-time Intervention, and many other mechanistic intervention technique. Here is what he says:
  
  pyvene  pushes for streamlining the sharing process of inference-time interventions and many more, comparing with other also super useful tools in this area!
  
  I created the activation diff (~0.14MB) based on your shared LLaMA-2-7b-chat by taking the bias terms. And your honest-llama can now be loaded as,
  
  ```python
  import torch
  from transformers import AutoTokenizer, AutoModelForCausalLM
  import pyvene as pv
  
  tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-2-7b-chat-hf")
  model = AutoModelForCausalLM.from_pretrained(
      "meta-llama/Llama-2-7b-chat-hf",
      torch_d

  COPY-PASTE FIX

  Copy

  # Honest LLaMA: Inference-Time Intervention for Eliciting Truthful Answers
  
  This repository provides an implementation of Inference-Time Intervention (ITI) to enhance the truthfulness and alignment of large language models, specifically LLaMA models, without requiring fine-tuning. It offers tools and replicated results for applying mechanistic interventions directly to model activations, making it valuable for AI researchers and developers focused on LLM safety and reliability.
  
  ### Update 08/24/2024
  With the release of LLaMA-3 models, I decided to replicate ITI on a suite of LLaMA models for easy comparison. I've recorded the results in `iti_replication_results.md` and uploaded the ITI baked-in models to HuggingFace here. Note that the ITI baked-in models and ITI applied to base models is not exactly a one-to-one comparison due to slight differences in when the activations are edited. The ITI baked-in models have the activation differences hardcoded into their attention biases. For more precise editing, consider only using the models' attention biases when processing tokens after the input prompt, to be more faithful to the original ITI method.
  
  -- Justin Ji @jujipotle
  
  ### Update 01/26/2024 :fire::fire:
  
  Zen provided this really cool library called pyvene that can be used to load Inference-time Intervention, and many other mechanistic intervention technique. Here is what he says:
  
  pyvene  pushes for streamlining the sharing process of inference-time interventions and many more, comparing with other also super useful tools in this area!
  
  I created the activation diff (~0.14MB) based on your shared LLaMA-2-7b-chat by taking the bias terms. And your honest-llama can now be loaded as,
  
  ```python
  import torch
  from transformers import AutoTokenizer, AutoModelForCausalLM
  import pyvene as pv
  
  tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-2-7b-chat-hf")
  model = AutoModelForCausalLM.from_pretrained(
      "meta-llama/Llama-2-7b-chat-hf",
      torch_d

  OpenDoneDismiss
• mediumabout#2

  Add a homepage URL to the repository's 'About' section

  Why:

  COPY-PASTE FIX

  Copy

  https://huggingface.co/likenneth/honest_llama (or a link to the associated paper/project page if available)

  OpenDoneDismiss

Category GEO backends resolved for this scan: google/gemini-2.5-flash, deepseek/deepseek-v4-flash

• CATEGORY QUERY

  How to improve language model truthfulness during inference without fine-tuning?

  you: not recommended
  AI recommended (in order):

  1. OpenAI's GPT-4
  2. Anthropic's Claude 3
  3. Google's Gemini
  4. LangChain
  5. LlamaIndex
  6. Google Search API
  7. Wikipedia API
  8. OpenAI's GPT-4 Turbo
  9. Anthropic's Claude 3 Opus
  10. Cohere's Command R+
  11. Wolfram Alpha API
  12. PubMed API
  13. Anthropic's Claude
  14. OpenAI's GPT-3.5
  15. Google's Gemini Pro
  16. Meta's Llama 2
  17. Mistral AI's Mixtral

  AI recommended 17 alternatives but never named likenneth/honest_llama. This is the gap to close.

  Show full AI answer
• CATEGORY QUERY

  Tools for applying mechanistic interventions to large language model activations?

  you: not recommended
  AI recommended (in order):

  1. TransformerLens
  2. Neuroscope
  3. CircuitsVis
  4. PyTorch
  5. Hugging Face Transformers
  6. Captum

  AI recommended 6 alternatives but never named likenneth/honest_llama. This is the gap to close.

  Show full AI answer

• Metadata completeness

  warn

  Suggestion:

• README presence

  pass

• Compared to common alternatives in this category, what is the core differentiator of likenneth/honest_llama?

  pass
  ⚠AI did not name likenneth/honest_llama — likely talking about a different project

  AI answers can be confidently wrong. Read for accuracy: does it match your actual tech stack, audience, and differentiator?

• If a team adopts likenneth/honest_llama in production, what risks or prerequisites should they evaluate first?

  pass
  ✓AI named likenneth/honest_llama explicitly

  AI answers can be confidently wrong. Read for accuracy: does it match your actual tech stack, audience, and differentiator?

• In one sentence, what problem does the repo likenneth/honest_llama solve, and who is the primary audience?

  pass
  ✓AI named likenneth/honest_llama explicitly

  AI answers can be confidently wrong. Read for accuracy: does it match your actual tech stack, audience, and differentiator?