SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2502.04384 · NeurIPS · 2025

Enhancing Reasoning to Adapt Large Language Models for Domain-Specific Applications

Xin Zhang, Bo Wen

arXiv · PDF · Open in the Atlas

Code that ran

1 of 3 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 1 failure of our instrument; 2 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2502.04384.json · badge: /api/ran/2502.04384.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
wenboown/generative-ai-for-semiconductor-physical-design canonical (classifier label)
classifier basis: classifier_2025
1 ran · 2 no recorded run
FunctionStatusWhere it lives
generate_latex_image_table Ran
ran; our contract draft was wrong, not the code
wenboown/generative-ai-for-semiconductor-physical-design/gen_visual_plots.py
code served (permissive licence) · get_code("4b6e922b10a023cc")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os
from collections import defaultdict

# Construct valid arguments for generate_latex_image_table

# image_paths: dict mapping shape names to llm data
image_paths = {
    'shape1.png': {
        'gpt-4o': {
            'pool': {'pool_all': 'path/to/gpt4o_pool.png'},
            'run': {1: 'path/to/gpt4o_run1.png', 2: 'path/to/gpt4o_run2.png'}
        },
        'claude-3-5-sonnet-20240620': {
            'pool': {'pool_all': 'path/to/claude_pool.png'},
            'run': {1: 'path/to/claude_run1.png', 2: 'path/to/claude_run2.png'}
        },
        'watsonx_meta-llama_llama-3-1-70b-instruct': {
            'pool': {'pool_all': 'path/to/llama70b_pool.png'},
            'run': {1: 'path/to/llama70b_run1.png', 2: 'path/to/llama70b_run2.png'}
        },
        'watsonx_meta-llama_llama-3-405b-instruct': {
            'pool': {'pool_all': 'path/to/llama405b_pool.png'},
            'run': {1: 'path/to/llama405b_run1.png', 2: 'path/to/llama405b_run2.png'}
        },
        'o1-preview': {
            'pool': {'pool_all': 'path/to/o1_pool.png'},
            'run': {1: 'path/to/o1_run1.png', 2: 'path/to/o1_run2.png'}
        }
    },
    'shape2.png': {
        'gpt-4o': {
            'pool': {'pool_all': 'path/to/gpt4o_pool2.png'},
            'run': {1: 'path/to/gpt4o_run1_2.png'}
        },
        'claude-3-5-sonnet-20240620': {
            'pool': {'pool_all': 'path/to/claude_pool2.png'},
            'run': {1: 'path/to/claude_run1_2.png'}
        },
        'watsonx_meta-llama_llama-3-1-70b-instruct': {
            'pool': {'pool_all': 'path/to/llama70b_pool2.png'},
            'run': {1: 'path/to/llama70b_run1_2.png'}
        },
        'watsonx_meta-llama_llama-3-405b-instruct': {
            'pool': {'pool_all': 'path/to/llama405b_pool2.png'},
            'run': {1: 'path/to/llama405b_run1_2.png'}
        },
        'o1-preview': {
            'pool': {'pool_all': 'path/to/o1_pool2.png'},
            'run': {1: 'path/to/o1_run1_2.png'}
        }
    }
}

# llms: list of LLM model names
llms = [
    'gpt-4o',
    'claude-3-5-sonnet-20240620',
    'watsonx_meta-llama_llama-3-1-70b-instruct',
    'watsonx_meta-llama_llama-3-405b-instruct',
    'o1-preview'
]

# runs: list of run identifiers
runs = [1, 2]

# groups: dict mapping group names to lists of shapes
groups = {
    'group_a': ['shape1', 'shape2'],
    'group_b': ['shape3']
}

# pool_experiments: list of pool experiment names (not used in function body, but required parameter)
pool_experiments = ['pool_all', 'pool_subset']

# questions: list of dicts with 'context' and 'question' keys
questions = [
    {'context': 'shape1.gds', 'question': 'What is the purpose of this circuit design?'},
    {'context': 'shape2.gds', 'question': 'How does this layout optimize for performance?'},
    {'context': 'shape3.gds', 'question': 'What are the key features of this design?'}
]

ARGS = {
    'image_paths': image_paths,
    'llms': llms,
    'runs': runs,
    'groups': groups,
    'pool_experiments': pool_experiments,
    'questions': questions
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
yaml — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
generate_question_tables Not yet run
unverified (harvested, no recorded run)
wenboown/generative-ai-for-semiconductor-physical-design/gen_visual_plots.py
code served (permissive licence) · get_code("5c368a6b045ad723")
How it ran

no invocation recorded. this sample carries JSON test cases instead of a setup block; get_reference_implementation serves them with the level they back.

latex_escape Not yet run
unverified (harvested, no recorded run)
wenboown/generative-ai-for-semiconductor-physical-design/gen_visual_plots.py
code served (permissive licence) · get_code("74b15cf0d0942bad")
How it ran

no invocation recorded. this sample carries JSON test cases instead of a setup block; get_reference_implementation serves them with the level they back.

Also present in other papers' code

Of 3 samples lifted from this paper's linked repositories, 3 carry a function identity and 0 do not. Of the 3 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 3 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

This paper presents SOLOMON, a novel Neuro-inspired Large Language Model (LLM) Reasoning Network architecture that enhances the adaptability of foundation models for domain-specific applications. Through a case study in semiconductor layout design, we demonstrate how SOLOMON enables swift adaptation of generalpurpose LLMs to specialized tasks by leveraging Prompt Engineering and In-Context Learning techniques. Our experiments reveal the challenges LLMs face in spatial reasoning and applying domain knowledge to practical problems. Results show that SOLOMON instances significantly outperform their baseline LLM counterparts and achieve performance comparable to state-of-the-art reasoning model, o1-preview. We discuss future research directions for developing more adaptive AI systems that can continually learn, adapt, and evolve in response to new information and changing requirements. The rapid advancements in large language models (LLMs) have revolutionized various aspects of artificial intelligence, enabling them to understand and generate human-like text with remarkable proficiency. However, adapting these general-purpose models to domain-specific tasks remains a significant challenge. In this paper, we introduce SOLOMON (System for Optimizing Language Outputs through Multi-agent Oversight Networks), a Neuro-inspired LLM Reasoning Network Architecture that leverages Prompt Engineering and In-Context Learning techniques, and demonstrate how SOLOMON can effectively adapt from its original design purpose in medical applications to a new domain: semiconductor layout design. Section 2 presents the SOLOMON architecture and highlights its design principles that contribute to enhanced adaptability. To provide context for our experiment, we first examine how a designer might attempt to use ChatGPT (with GPT-4o mode) for a via connection design task in section 3. This exploration reveals a critical limitation: while LLMs can accurately recite textbook definitions of domain-specific concepts, they struggle to extract and apply expert knowledge to solve practical tasks. Human needs to translate high-level concepts into specific geometric requirements, which the LLM can then use to generate code for drawing shapes. This highlights the key challenge in adapting LLMs for domain-specific applications: their limited reasoning capabilities. In section 4, we developed a set of 25 tasks ranging from basic geometric shapes to complex semiconductor structures, to evaluate our SOLOMON architecture against five different LLMs. These tasks assess spatial reasoning capabilities and adaptability across various complexity levels. Through these experiments, we demonstrate SOLOMON's superior performance compared to standalone LLMs, and reaching the level of state-of-the-art reasoning models like O1-preview. 38th Conference on Neural Information Processing Systems (NeurIPS 2024).

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2502.04384")
get_code_for_paper("2502.04384")
have("2502.04384")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2502.04384.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2502.04384.svg)](https://syntology.ai/paper/2502.04384)

Connect an agent — have() is free.