datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
github-codereview
Code Review Dataset
A large-scale dataset of the best human-written code reviews from top GitHub repositories.
Each row captures a moment where a human code reviewer left an inline comment on a pull request, and the author subsequently modified the code in response.
The dataset also includes negative examples — code from the same PRs that passed review without comments — to help models learn when code is acceptable.
This provides a natural signal for training models to:
Generate… See the full description on the dataset page: https://huggingface.co/datasets/ronantakizawa/github-codereview.code-review-bench
Code Review Bench
A paired online-offline benchmark for AI code review.
Splits
online — Stratified sample of 1,135 bot-reviewed PRs, scraped from open-source Github repositories and scored by the online benchmark (15 tools, Feb–Apr 2026).
offline — 136 expert-curated golden issues across 50 PRs (5 repositories).
Provenance
The offline golden issues extend the 50-PR benchmark originally created by Greptile (2025) and refined by Augment (2025). Our… See the full description on the dataset page: https://huggingface.co/datasets/code-review-bench/code-review-bench.codereviewergemma4-code-review-instruct
gemma4-code-review-instruct
197K code review examples — 58K with chain-of-thought <think> reasoning traces.
Built to train models that don't just flag issues, but explain their reasoning before delivering a review. Drop-in ready for SFT with any chat model.
Why This Dataset
Most code review datasets give you diff → comment. This one gives you diff → think → comment for 30% of examples — reasoning traces that show how to analyze a diff before writing the review.… See the full description on the dataset page: https://huggingface.co/datasets/liodon-ai/gemma4-code-review-instruct.code-review
CODE_REVIEW
A preference dataset for CODE_REVIEW, harvested from real, human-labelled sources and curated by an automated harvesting harness with an LLM quality gate.
Format
Standard preference / DPO schema — each row:
column
meaning
prompt
the request (originally code)
chosen
the human-preferred response
rejected
a worse response to the same prompt
source
the dataset/URL the row was harvested from
Splits
80/10/10 train /… See the full description on the dataset page: https://huggingface.co/datasets/316usman/code-review.code-review-instruct-critique-revision
Dataset Card for "code-review-instruct-critique-revision"
More Information needed
github-codereview-dataset
Github-Codereview-Dataset
Made with ❤️ using 🦥 Unsloth Studio
github-codereview-dataset was generated with Unsloth Recipe Studio. It contains 10,000 generated records.
🚀 Quick Start
from datasets import load_dataset
# Load the main dataset
dataset = load_dataset("manishsaini1/github-codereview-dataset", "data", split="train")
df = dataset.to_pandas()
📊 Dataset Summary
📈 Records: 10,000
📋 Columns: 23
📋 Schema & Statistics… See the full description on the dataset page: https://huggingface.co/datasets/manishsaini1/github-codereview-dataset.dl4se-code-review-datasetmultilang-code-quality-reviewscodereview-bench
CodeReview-Bench
A benchmark for evaluating models on two code review tasks, curated from ronantakizawa/github-codereview.
Tasks
1. Code Editing
Given code and a reviewer comment, apply the requested change.
Input: before_code, reviewer_comment, language, diff_context
Target: after_code
from datasets import load_dataset
ds = load_dataset("ronantakizawa/codereview-bench", "code-editing")
example = ds["test"][0]
prompt = f"""Apply the following review comment… See the full description on the dataset page: https://huggingface.co/datasets/ronantakizawa/codereview-bench.gordon-ramsay-code-review-v2
Gordon Ramsay Code Review & Auditor Corpus v2 (dcmutlu/gordon-ramsay-code-review-v2)
A high-density synthetic dataset of 10,000 multi-turn code review pairs designed to fine-tune open-weight reasoners (specifically Qwen2.5-Coder-7B-Instruct) into Chef Gordon Ramsay: Sovereign Executive Code Auditor and Supreme Software Gastronomer.
🍳 Dataset Overview
This dataset merges rigorous computer science diagnostics (Abstract Syntax Tree inspection, concurrency lifecycle… See the full description on the dataset page: https://huggingface.co/datasets/dcmutlu/gordon-ramsay-code-review-v2.code-reviewA Scrape of the codereview stack exchange, good for high quality code
stackexchange-codereview-sandboxescode-review-instruct-critique-revision-pythoncode-review-benchmarkagent-traces-code-review-pipeline
Agent Traces: code-review-pipeline
Synthetic multi-agent workflow traces with LLM-enriched content for the code-review-pipeline domain.
Part of the juliensimon/open-agent-traces collection — 10 datasets covering diverse domains and workflow patterns.
What is this dataset?
This dataset contains 2,035 events across 50 workflow runs, each representing a complete multi-agent execution trace. Every trace includes:
Agent reasoning — chain-of-thought for each agent step
LLM… See the full description on the dataset page: https://huggingface.co/datasets/juliensimon/agent-traces-code-review-pipeline.stackexchange_codereviewswebench-codereview-benchmark-v3
SWE-bench Code Review Benchmark v3
This dataset contains 7 benchmark splits for evaluating code review models on the SWE-bench task.
Dataset Summary
Total instances: 3500
Total resolved: 801 (22.9%)
Splits: 7 (3 main + 4 weak models)
Version: 3.0.0
Created: 2026-05-03
Splits
Split
Instances
Resolved
Resolve Rate
Model
glm5_500_v3
500
361
72.2%
openai/GLM-5-FP8
qwen3_coder_30b_500_v3
500
235
47.0%
Qwen/Qwen3-Coder-30B-A3B-Instruct… See the full description on the dataset page: https://huggingface.co/datasets/ruoyu001/swebench-codereview-benchmark-v3.korean_code_reviews_from_githubbase_code_review
Dataset Card for "base_code_review"
More Information needed
react_native_code_review-reasoning-SFTCodeReviewQA CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models
The task of automated code refinement aims to automate the developer's perspective in resolving an actionable code review comment provided by a reviewer.
This is a generative task, where the LLM is required to revise a pre-review code submission with respect to the natural language code review comment to produce an intended post-review code revision.
CodeReviewQA further breaks down this generative task… See the full description on the dataset page: https://huggingface.co/datasets/Tomo-Melb/CodeReviewQA.amazon-reviews-for-llm-extended
Cross-domain sequential recommendation dataset
A sequential recommendation dataset drawn from Amazon Reviews 2023, covering
Books, CDs_and_Vinyl, Movies_and_TV, Video_Games.
Each row of interactions.parquet is one user buying or reviewing one item at one time.
Users are sampled so that every one of them is active in all domains, their
interactions are ordered chronologically and cut into train/valid/test, and each
interaction carries a fixed set of 10 candidate items for ranking… See the full description on the dataset page: https://huggingface.co/datasets/sungjin-code/amazon-reviews-for-llm-extended.reasoning-sft-github-codereview
reasoning-sft-github-codereview
Converted version of ronantakizawa/github-codereview, filtered to 76,689 high-quality rows (quality_score >= 0.75, excluding none comment type).
Nothing fancy, just reformatted the columns into a standard messages format for SFT/reasoning training. No content was modified or regenerated.
Format
Each row has three columns:
input — list of dicts with role and content (system prompt + user turn containing the reviewer comment and original… See the full description on the dataset page: https://huggingface.co/datasets/AmanPriyanshu/reasoning-sft-github-codereview.amazon-reviews-books-for-llm
Books sequential recommendation dataset
A sequential recommendation dataset drawn from Amazon Reviews 2023, covering
Books.
Each row of interactions.parquet is one user buying or reviewing one item at one time.
Users are sampled so that every one of them is active in all domains, their
interactions are ordered chronologically and cut into train/valid/test, and each
interaction carries a fixed set of 10 candidate items for ranking
evaluation. Integer user_idx / item_idx columns… See the full description on the dataset page: https://huggingface.co/datasets/sungjin-code/amazon-reviews-books-for-llm.amazon-reviews-for-llm
Cross-domain sequential recommendation dataset
A sequential recommendation dataset drawn from Amazon Reviews 2023, covering
Books, CDs_and_Vinyl, Movies_and_TV, Video_Games.
Each row of interactions.parquet is one user buying or reviewing one item at one time.
Users are sampled so that every one of them is active in all domains, their
interactions are ordered chronologically and cut into train/valid/test, and each
interaction carries a fixed set of 10 candidate items for ranking… See the full description on the dataset page: https://huggingface.co/datasets/sungjin-code/amazon-reviews-for-llm.code-review-qwen35
Code Review Corpus
Description
Teaches domain-specific instruction following and code generation for this expert.
Source
ronantakizawa/github-codereview - Other
code-review-bench/code-review-bench - CC-BY-4.0
Formatted for the MoE-orchestrator project
(https://github.com/michaelowusuntim6/MoE-orchestrator). Expert target:
debug_review.
Format
Each record is a JSON object with a messages field formatted for Qwen3.5's
native chat… See the full description on the dataset page: https://huggingface.co/datasets/michaelowusuntim6/code-review-qwen35.2048_has_code_filtered_base_code_review_python_based_on_property
Dataset Card for "2048_has_code_filtered_base_code_review_python_based_on_property"
More Information needed
big-multilang-code-quality-reviewsterminal_bench_2_stackexchange_codereview_sandboxes_traces_terminus_2_overwrite6d215e47