datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
ParseBench
ParseBench
Quick links: [🌐 Website] [📜 Paper] [💻 Code]
ParseBench is a benchmark for evaluating document parsing systems on real-world enterprise documents, with the following characteristics:
Multi-dimensional evaluation. The benchmark is stratified into five capability dimensions — tables, charts, content faithfulness, semantic formatting, and visual grounding — each with task-specific metrics designed to capture what agentic workflows depend on.
Real-world enterprise… See the full description on the dataset page: https://huggingface.co/datasets/llamaindex/ParseBench.ParseBench
ParseBench
Quick links: [🌐 Website] [📜 Paper] [💻 Code]
ParseBench is a benchmark for evaluating document parsing systems on real-world enterprise documents, with the following characteristics:
Multi-dimensional evaluation. The benchmark is stratified into five capability dimensions — tables, charts, content faithfulness, semantic formatting, and visual grounding — each with task-specific metrics designed to capture what agentic workflows depend on.
Real-world enterprise… See the full description on the dataset page: https://huggingface.co/datasets/Yoongls/ParseBench.ParseBench
ParseBench
Quick links: [🌐 Website] [📜 Paper] [💻 Code]
ParseBench is a benchmark for evaluating document parsing systems on real-world enterprise documents, with the following characteristics:
Multi-dimensional evaluation. The benchmark is stratified into five capability dimensions — tables, charts, content faithfulness, semantic formatting, and visual grounding — each with task-specific metrics designed to capture what agentic workflows depend on.
Real-world enterprise… See the full description on the dataset page: https://huggingface.co/datasets/huggingworld/ParseBench.parsed_dataParseBench
ParseBench
Quick links: [🌐 Website] [📜 Paper] [💻 Code]
ParseBench is a benchmark for evaluating document parsing systems on real-world enterprise documents, with the following characteristics:
Multi-dimensional evaluation. The benchmark is stratified into five capability dimensions — tables, charts, content faithfulness, semantic formatting, and visual grounding — each with task-specific metrics designed to capture what agentic workflows depend on.
Real-world enterprise… See the full description on the dataset page: https://huggingface.co/datasets/matthew-l-leidos/ParseBench.ParseBench_test
ParseBench
Quick links: [🌐 Website] [📜 Paper] [💻 Code]
ParseBench is a benchmark for evaluating document parsing systems on real-world enterprise documents, with the following characteristics:
Multi-dimensional evaluation. The benchmark is stratified into five capability dimensions — tables, charts, content faithfulness, semantic formatting, and visual grounding — each with task-specific metrics designed to capture what agentic workflows depend on.
Real-world enterprise… See the full description on the dataset page: https://huggingface.co/datasets/kramp/ParseBench_test.pdf-parse-bench
PDF Parse Bench
Benchmark for evaluating how effectively PDF parsing solutions extract mathematical formulas and tables from documents.
We generate synthetic PDFs with diverse formatting scenarios, parse them with different parsers, and score the extracted content using LLM-as-a-Judge. This semantic evaluation approach substantially outperforms traditional metrics in agreement with human judgment.
Leaderboard (2026-Q1)
Results are based on two benchmark… See the full description on the dataset page: https://huggingface.co/datasets/piushorn/pdf-parse-bench.metamon-parsed-replays
Metamon Replay Dataset
Pokémon Showdown replay files parsed (or "reconstructed") into RL trajectories by Metamon (arXiv Appendix D)
Quick Start
The easiest way to use the replay dataset is through metamon's dataloader:
import metamon
from metamon.interface import get_observation_space, get_reward_function, get_action_space
from metamon.data importParsedReplayDataset
# see the metamon README for more on observations, actions, and rewards.
human_dset =… See the full description on the dataset page: https://huggingface.co/datasets/jakegrigsby/metamon-parsed-replays.pdf-parse-bench
PDF Parse Bench
Benchmark for evaluating how effectively PDF parsing solutions extract mathematical formulas and tables from documents.
We generate synthetic PDFs with diverse formatting scenarios, parse them with different parsers, and score the extracted content using LLM-as-a-Judge. This semantic evaluation approach substantially outperforms traditional metrics in agreement with human judgment.
Leaderboard (2026-Q1)
Results are based on two benchmark… See the full description on the dataset page: https://huggingface.co/datasets/ankitt6174/pdf-parse-bench.ParserV1-modelsagenda-parser-tool-traces
Agenda Parser — tool-calling reasoning traces
ReAct tool-calling traces for the Agenda Parser
agents: each row is one agent step — a {system, user, assistant} chat example
where the assistant emits a single JSON action {"thought", "tool", "args"}.
Two agents are covered (tagged by meta.domain):
agenda — the uploaded-packet research agent, over real public-meeting agenda
packets (tools: list/read items, semantic + exact search, summarize, report).
Each agenda row's meta.unit_id… See the full description on the dataset page: https://huggingface.co/datasets/build-small-hackathon/agenda-parser-tool-traces.rosettacode-parsed
Data Origins
Original dataset: https://huggingface.co/datasets/jondurbin/rosettacode-raw/
Cleaner code: https://github.com/the-crypt-keeper/rosettacode-parser
Data Fields
Field
Type
Description
title
string
problem title
task
string
problem description
language
string
solution language/variant
soulution
string
solution source code
Languages
One .jsonl is provided per language group, the sublanguage field in the data denotes the… See the full description on the dataset page: https://huggingface.co/datasets/mike-ravkine/rosettacode-parsed.parsebench-table-track
ParseBench Table Track plus Financial Split
This dataset is a mirror of the table dimension of llamaindex/ParseBench, packaged together with a curated Financial Split that we built for evaluating OCR systems on insurance and financial filings.
It contains,
All 503 PDFs of the ParseBench table track.
table.jsonl, the original ground truth (one HTML table per page, plus easy or hard difficulty tag).
financial_split/, our 151 page financial slice plus the 117 dropped non financial… See the full description on the dataset page: https://huggingface.co/datasets/roma2025/parsebench-table-track.chart-parse-bench
ChartParse-Bench
58 synthetic charts whose exact values were written before the pixels existed.
Every gate in chart extraction measures self-consistency: redraw the extraction, check it
lands on the same ink, call it verified. That passes a series traced confidently off the
wrong axis. Because each chart here ships the array it was drawn from, the question stops
being is this answer plausible and becomes is this answer right.
Contents
family
charts
series… See the full description on the dataset page: https://huggingface.co/datasets/Unsiloed/chart-parse-bench.glaive-function-calling-v2-ru-parsed
Glaive Function Calling V2 Parsed (RU)
This dataset is a fork of an original mizinovmv/glaive-function-calling-v2-ru.
Structure
train (102860 rows)
messages
functions
test (9681 rows)
messages
functions
Messages field contains JSON string with list of messages. Each message has role (one of: system, user, assistant, function_call, function_response) and content (message string) fields.
Functions field contains JSON string with list of functions for current chat… See the full description on the dataset page: https://huggingface.co/datasets/korotkov/glaive-function-calling-v2-ru-parsed.parsebench-recipe-runs
ParseBench recipe runs
Experiment log for document-parsing runs on the public ParseBench test
subset (llamaindex/ParseBench), scored with the official open-source
ParseBench evaluator (run-llama/ParseBench).
Recipes combine CLI coding agents doing vision parsing with deterministic
PDF text-layer tools (word-bbox snapping, style extraction from span flags
and vector-drawing geometry).
Sample: official test subset - 12 single-page PDFs, 3 per category
(chart / layout / table /… See the full description on the dataset page: https://huggingface.co/datasets/sleepyheeler/parsebench-recipe-runs.pi-trace-parser-sessionsGRPO-Lambda-ParsedForUnsloth
This is a dataset I've generated containing about 50,000 JSON datapoints of the lambda calculus, including prompts for the problem, and extended explanation of the solution process. It was originally generated in pure JSON with extra metadata for myself. There are 50 separate files ranging on average 2.7 MB, each containing 1000 datapoints, this is split for the at home trainers with low GPU. This dataset is converted to a Question-Reasoning-Answer style ideally (hopefully) for use with… See the full description on the dataset page: https://huggingface.co/datasets/Creekside/GRPO-Lambda-ParsedForUnsloth.ParseEmbed
ParseEmbed
Hard, parse-sensitive retrieval evaluation for embedding models.
ParseEmbed is a compact benchmark for embedding models. It tests whether a
model can retrieve the exact correct document when hard negatives share nearly
all surface tokens with the answer.
Tasks
Task ID
Split
What it measures
mean
mean
Semantic scope, negation, numeric values, temporal conditions, and exception handling
text_formatting
text_formatting
Meaning carried by Markdown-like… See the full description on the dataset page: https://huggingface.co/datasets/Convence/ParseEmbed.glaive-function-calling-v2-parsed
Glaive Function Calling V2 Parsed
This dataset is a fork of an original - glaiveai/glaive-function-calling-v2.
Structure
train (100199 rows)
messages
functions
test (12553 rows)
messages
functions
Messages field contains JSON string with list of messages. Each message has role (one of: system, user, assistant, function_call, function_response) and content (message string) fields.
Functions field contains JSON string with list of functions for current chat (their… See the full description on the dataset page: https://huggingface.co/datasets/korotkov/glaive-function-calling-v2-parsed.agenda-parser-tool-traces
Agenda Parser — tool-calling reasoning traces
ReAct tool-calling traces for the Agenda Parser
agents: each row is one agent step — a {system, user, assistant} chat example
where the assistant emits a single JSON action {"thought", "tool", "args"}.
Two agents are covered (tagged by meta.domain):
agenda — the uploaded-packet research agent, over real public-meeting agenda
packets (tools: list/read items, semantic + exact search, summarize, report).
Each agenda row's meta.unit_id… See the full description on the dataset page: https://huggingface.co/datasets/rdubwiley/agenda-parser-tool-traces.Item-Parser-Dataset
Contents:
~$0.80 API token usage for Gemini 2.0 Flash Lite
OpenMathInstruct-1-ParsedThis is a parsed subset of the OpenMathInstruct-1 dataset.
Each sample in the dataset is formatted in the following way:
{"messages": [
{"role": "user", "content": "user prompt"},
{"role": "assistant", "content": "assistant response", "tool_call": "python code"},
{"role": "tool", "content": "code execution result"},
{"role": "assistant","content": "assistant response"},
]}
The dataset is split into two files:
open-math-short.jsonl - the assistant only makes a single function call.… See the full description on the dataset page: https://huggingface.co/datasets/kaleinaNyan/OpenMathInstruct-1-Parsed.grok-parser-vrl-940k
grok-parser-vrl-940k
940,257 validated (log, grok_pattern) pairs for training models that
generate Vector.dev VRL parse_grok! patterns from raw log lines.
Files
merged_validated.csv — full schema (854 MB)
log — raw log line
parser — full VRL snippet (e.g. .message = ... | parse_grok!(.message, "..."))
grok_pattern — bare grok string extracted from parser
target — canonical pygrok output (dict)
parsed_output — independent re-application of grok_pattern (sanity check)… See the full description on the dataset page: https://huggingface.co/datasets/omeryentur/grok-parser-vrl-940k.sql-parsedmath_trajectory_parsed
MATH Trajectory Parsed
Overview
This dataset contains DeepSeek-R1 reasoning trajectories for MATH-style problems, each decomposed into a sequence of labeled reasoning steps. Every trajectory is provided at two levels of granularity in parallel: the full-length text of each step (steps), and a condensed summary of each step (step summaries), with a short descriptor (step descriptors) linking the two.
The dataset has two problem sources:
Config
Rows
Problem… See the full description on the dataset page: https://huggingface.co/datasets/sxiong/math_trajectory_parsed.agenda-parser-models-example-agent-traces
Agenda Parser — fine-tuned agent models
Three Gemma 4 models fine-tuned to drive the Agenda Parser's ReAct agent: at each step
the model emits a single JSON action {"thought","tool","args"} over two toolkits —
meeting-agenda packets and Michigan local-government law (Open Meetings Act, FOIA,
the Michigan Compiled Laws via Cornell LII). This card doubles as the project write-up; the
dataset itself (bottom) is a gallery of example traces from the three models.
tier
base… See the full description on the dataset page: https://huggingface.co/datasets/build-small-hackathon/agenda-parser-models-example-agent-traces.Item-Parser-Dataset-v1.1-1k
Contents:
dataset with varying items parsed instead of parsing items from the same category. data generated by Qwen3 32B for first ~150 chats,
remaining 700+ listing and chat parsed outputs are generated by Qwen3 30B A3B, total API usage is ~US$0.70
metamon-parsed-wm-replays
Metamon Parsed World-Model Replays
This dataset contains Metamon replay parser text outputs for world-model
training. These files are produced by python -m metamon.backend.replay_parser
and are consumed by Metamon tokenizer and world-model dataset generation tools.
Storage layout
archives/<format>.tar.gz: parsed replay .txt files for one battle format.
indexes/<format>.jsonl.gz: one row per replay text file with path,
size_bytes, and sha256.… See the full description on the dataset page: https://huggingface.co/datasets/sooham34/metamon-parsed-wm-replays.pickai-synthetic-nl-parse-v1
PickAI Synthetic NL Parse v1
Synthetic instruction dataset for mapping warehouse supervisor natural language into PickAI OptimizeConstraints JSON. Each row pairs an operator-style instruction with ground-truth constraint fields produced by the deterministic PickAI optimizer, not hand-labeled guesses.
Part of PickAI, an open-source WMS-adjacent pick-path optimization service.
Related model (experimental LoRA, value gate failed): MuhibBeekun/pickai-qwen2.5-7b-nl-parse-lora… See the full description on the dataset page: https://huggingface.co/datasets/MuhibBeekun/pickai-synthetic-nl-parse-v1.