{
  "id": "2311.01460",
  "title": "Implicit Chain of Thought Reasoning via Knowledge Distillation",
  "related_terms": [
    "implicit CoT",
    "latent reasoning",
    "latent chain-of-thought",
    "latent CoT"
  ],
  "authors": "Yuntian Deng, Kiran Prasad, Roland Fernandez, Paul Smolensky, Vishrav Chaudhary, Stuart Shieber",
  "first_public": "2023-11-02",
  "venue": "arXiv",
  "arxiv": "2311.01460",
  "paper_url": "https://arxiv.org/abs/2311.01460",
  "code_url": "https://github.com/da03/implicit_chain_of_thought/",
  "lens": [
    "internalized"
  ],
  "representation": [
    "hidden-state"
  ],
  "computation": [
    "depth-wise"
  ],
  "training": [
    "distillation"
  ],
  "questions": [
    "capability",
    "efficiency"
  ],
  "evidence": [
    "empirical"
  ],
  "modality": [
    "language",
    "symbolic"
  ],
  "summary": "First proposes implicit chain-of-thought reasoning: language models carry out intermediate reasoning in continuous hidden states rather than generate it as text. Defines this setting separately from its implementation, which distills an explicit-CoT teacher's hidden states into an emulator and a student that produces the answer.",
  "provenance": "Awesome-Latent-CoT",
  "author_names": [
    "Yuntian Deng",
    "Kiran Prasad",
    "Roland Fernandez",
    "Paul Smolensky",
    "Vishrav Chaudhary",
    "Stuart Shieber"
  ],
  "record_url": "https://latent-implicit-thinking.github.io/papers/2311.01460/",
  "legacy_record_url": "https://latent-implicit-thinking.github.io/papers/#paper-2311-01460",
  "citation_date": "2023-11-02",
  "json_url": "https://latent-implicit-thinking.github.io/papers/2311.01460/record.json",
  "markdown_url": "https://latent-implicit-thinking.github.io/papers/2311.01460/index.md",
  "bibtex_url": "https://latent-implicit-thinking.github.io/papers/2311.01460/citation.bib",
  "csl_url": "https://latent-implicit-thinking.github.io/papers/2311.01460/citation.csl.json",
  "demos": [
    {
      "id": "multiplication",
      "title": "Multiplication without written steps",
      "label": "Multiplication",
      "description": "Compare GPT-2 models using implicit CoT, explicit CoT, or no CoT on two numbers of up to 20 digits each.",
      "limitations": "The models can make mistakes, including on valid inputs. The implicit model writes answer digits, not intermediate steps.",
      "model": "GPT-2",
      "url": "https://huggingface.co/spaces/yuntian-deng/gpt2-multiplication",
      "model_url": "https://huggingface.co/yuntian-deng/gpt2-implicit-cot-multiplication-20-digits",
      "topic_ids": [
        "internalized"
      ],
      "paper_relationships": [
        {
          "paper_id": "2405.14838",
          "relationship": "implementation"
        },
        {
          "paper_id": "2311.01460",
          "relationship": "related_method"
        }
      ],
      "relationship": "related_method"
    },
    {
      "id": "math-word-problems",
      "title": "Math word problems without written steps",
      "label": "Math word problems",
      "description": "Try a Mistral-7B model trained with stepwise internalization on grade-school math word problems.",
      "limitations": "Answers can be wrong. This model is trained for grade-school math word problems, not general conversation. It still generates the answer token by token.",
      "model": "Mistral-7B",
      "url": "https://huggingface.co/spaces/yuntian-deng/implicit-cot-math",
      "model_url": "https://huggingface.co/yuntian-deng/implicit-cot-math-mistral7b",
      "topic_ids": [
        "internalized"
      ],
      "paper_relationships": [
        {
          "paper_id": "2405.14838",
          "relationship": "implementation"
        },
        {
          "paper_id": "2311.01460",
          "relationship": "related_method"
        }
      ],
      "relationship": "related_method"
    }
  ],
  "method_comparisons": [
    {
      "title": "Distillation and stepwise internalization",
      "text": "Deng et al. (2023) introduced implicit CoT as reasoning through hidden states rather than written steps; distillation is the paper's implementation of that proposal. It learns to predict a teacher's intermediate hidden states, which help a student produce the answer. Stepwise Internalization instead repeatedly shortens the written solution and fine-tunes the same model on what remains. Both use worked solutions during training; neither requires writing the removed steps at inference. Their reported experiments focus on multiplication and grade-school math, not a guarantee of general reasoning ability.",
      "source_ids": [
        "2311.01460",
        "2405.14838"
      ],
      "source_urls": [
        "https://arxiv.org/html/2311.01460v1#S2.SS3",
        "https://arxiv.org/html/2311.01460v1#S3",
        "https://arxiv.org/html/2405.14838v1#S3"
      ],
      "source_labels": [
        "Implicit CoT proposal, §2.3",
        "Distillation implementation, §3",
        "Stepwise internalization, §3"
      ],
      "topic_id": "internalized",
      "topic_label": "Internalized reasoning",
      "topic_url": "https://latent-implicit-thinking.github.io/topics/internalized-reasoning/"
    },
    {
      "title": "Hidden-state distillation: the 2023 method and CODI",
      "text": "The 2023 implicit CoT paper proposes reasoning through hidden states and implements it with a teacher, a thought emulator and an answer-generating student. CODI also transfers explicit reasoning through hidden-state supervision, but uses one model for both teacher and student tasks. It aligns their states just before the answer, while the student takes successive continuous reasoning steps.",
      "source_ids": [
        "2311.01460",
        "2502.21074"
      ],
      "source_urls": [
        "https://arxiv.org/html/2311.01460v1#S2.SS3",
        "https://arxiv.org/html/2311.01460v1#S3",
        "https://arxiv.org/html/2502.21074v3#S3"
      ],
      "source_labels": [
        "Deng et al. (2023), §2.3",
        "Deng et al. (2023), §3",
        "CODI, §3"
      ],
      "topic_id": "internalized",
      "topic_label": "Internalized reasoning",
      "topic_url": "https://latent-implicit-thinking.github.io/topics/internalized-reasoning/"
    },
    {
      "title": "Implicit CoT: proposal and implementation",
      "text": "Deng et al. (2023) proposed implicit chain-of-thought reasoning: language models carry out intermediate reasoning in continuous hidden states rather than generate it as text. Section 2.3 defines this setting; section 3 develops a distillation method to implement it. That method transfers a teacher's reasoning into hidden states across layers. Reasoning through continuous hidden states is the broader proposal; it does not require the particular distillation procedure used in this paper. Its across-layer computation differs from the repeated hidden-state feedback used by later methods such as Coconut.",
      "source_ids": [
        "2311.01460",
        "2412.06769"
      ],
      "source_urls": [
        "https://arxiv.org/html/2311.01460v1#S2.SS3",
        "https://arxiv.org/html/2311.01460v1#S3",
        "https://arxiv.org/html/2412.06769v3#S3"
      ],
      "source_labels": [
        "Implicit CoT proposal, §2.3",
        "Distillation implementation, §3",
        "Coconut, §3"
      ],
      "topic_id": "continuous",
      "topic_label": "Continuous-state reasoning",
      "topic_url": "https://latent-implicit-thinking.github.io/topics/continuous-states/"
    }
  ]
}