datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
Layer_GradientGradients_Gradients_and_Text_Full_Logic_CaptionsSSR-3DFRONT
SSR-3DFRONT: Structured Scene Representation for 3D Indoor Scenes
This dataset provides a processed version of the 3D-FRONT dataset with structured scene representations for text-driven 3D indoor scene synthesis and editing.
Mor information about ReSpace: http://respace.mnbucher.com
For detailed usage instructions, training details, and examples, see the associated repository: https://github.com/GradientSpaces/respace
Our model weights for SG-LLM:… See the full description on the dataset page: https://huggingface.co/datasets/gradient-spaces/SSR-3DFRONT.SmolLM2-135M-gradients-10Mdetails_gradientai__v-alpha-tross
Dataset Card for Evaluation run of gradientai/v-alpha-tross
Dataset automatically created during the evaluation run of model gradientai/v-alpha-tross on the Open LLM Leaderboard.
The dataset is composed of 63 configuration, each one coresponding to one of the evaluated task.
The dataset has been created from 2 run(s). Each run can be found as a specific split in each configuration, the split being named using the timestamp of the run.The "train" split is always pointing to the… See the full description on the dataset page: https://huggingface.co/datasets/open-llm-leaderboard-old/details_gradientai__v-alpha-tross.gradient_accumulation_exampleWild-SLAMThis repository contains data for WildGS-SLAM: Monocular Gaussian Splatting SLAM in Dynamic Environments.
Paper | Project Page | Code
WildGS-SLAM accurately tracks the camera trajectory and reconstructs a 3D Gaussian map for static elements from a monocular video sequence, effectively removing dynamic components.
Datasets Used
WildGS-SLAM uses data from the following datasets:
Wild-SLAM Mocap Dataset: (Hugging Face) Download instructions are available in the github repository.… See the full description on the dataset page: https://huggingface.co/datasets/gradient-spaces/Wild-SLAM.details_gradientai__Llama-3-70B-Instruct-Gradient-262kfineweb_synth_dense_ocr_gradients_with_grounding_Bfineweb_synth_dense_ocr_gradients_with_grounding_Adetails_diffnamehard__Mistral-CatMacaroni-slerp-gradient
Dataset Card for Evaluation run of diffnamehard/Mistral-CatMacaroni-slerp-gradient
Dataset automatically created during the evaluation run of model diffnamehard/Mistral-CatMacaroni-slerp-gradient on the Open LLM Leaderboard.
The dataset is composed of 63 configuration, each one coresponding to one of the evaluated task.
The dataset has been created from 1 run(s). Each run can be found as a specific split in each configuration, the split being named using the timestamp of the… See the full description on the dataset page: https://huggingface.co/datasets/open-llm-leaderboard-old/details_diffnamehard__Mistral-CatMacaroni-slerp-gradient.WildChat-4.8M-filtered-gpt-4SWE-ZERO-12M-trajectories-filtered
Filtered SWE Trajectories — 60K
This dataset contains 60,000 software-engineering agent trajectories from two
upstream datasets:
50,000 Submitted trajectories sampled from
AlienKevin/SWE-ZERO-12M-trajectories
10,000 accepted trajectories sampled from
Kwai-Klear/SWE-smith-mini_swe_agent_plus-trajectories-66k
All rows use this common schema:
instance_id
repo
messages
trajectory_format
exit_status
duration_sec
For the SWE-smith rows, repo was derived from the source instance_id… See the full description on the dataset page: https://huggingface.co/datasets/gradients-io-tournaments/SWE-ZERO-12M-trajectories-filtered.intercode_bigcode_combined_12k
Intercode Bigcode Combined 12K
Balanced instruction/response dataset with 12,000 examples:
6,000 examples sampled from the filtered Tellina NL2Bash-derived InterCode training pool
6,000 examples sampled from bigcode/self-oss-instruct-sc2-exec-filter-50k
The dataset file is intercode_bigcode_combined_12k.jsonl.
Each row has:
{"instruction": "...", "response": "..."}
Sampling metadata is stored in intercode_bigcode_combined_12k.meta.json.
code-gradients
Gradients code comparison
This public copy contains 18,718 training rows and 998 test rows. Each row has
exactly one user turn followed by one assistant turn under conversations. The local
source uses the equivalent top-level key messages; only that key was renamed for
compatibility with Gradients baseline preparation.
multigesture-mimic-green-gradientpvp-tool-calling-sft
PvP tool-calling SFT cold-start data
Claude-vs-Claude games played through the G.O.D PvP tool-calling harness. Each row is one model turn (or post-game reflection): the system+user prompt the harness built, the assistant response (content + tool_calls), and the tools schemas — i.e. the OpenAI messages+tools format consumed by tokenizer.apply_chat_template(messages, tools=tools). On a move turn the assistant co-emits any memory-tool edits and a game_action committing a legal… See the full description on the dataset page: https://huggingface.co/datasets/gradients-io-tournaments/pvp-tool-calling-sft.repro-chain-of-thought-gradient-descent-runs-sol
Chain-of-Thought Gradient Descent reproduction runs
Immutable outputs for the independent scaled reproduction of ICML 2026 paper
#443, OpenReview uZ8JZ1Lw9a.
gpu-l4-seed443/: successful NVIDIA L4 checkpoint, result JSON, and cost CSV.
figures/: interactive logbook figures and raw CSVs.
poster/: Posterly source, zero-warning gate report, PDF/PNG, and
self-contained poster_embed.html.
reproduction-bundle/: complete clean download-and-rerun bundle.
Successful Job:… See the full description on the dataset page: https://huggingface.co/datasets/JG1310/repro-chain-of-thought-gradient-descent-runs-sol.repro-chain-of-thought-gradient-descent-code
Reproduction bundle — Chain-of-Thought Gradient Descent (ICML 2026, submission #443)
OpenReview: https://openreview.net/forum?id=uZ8JZ1Lw9a
An independent reproduction. No official code, checkpoints, or data were released with
the paper, so everything here is implemented from the paper text (Definitions 2.2–2.3,
4.3–4.4; Theorems 4.1–4.2; Propositions 4.1–4.4; Appendix D and E).
What is here
scripts/
common.py N-layer FFN forward/backprop (Def… See the full description on the dataset page: https://huggingface.co/datasets/JG1310/repro-chain-of-thought-gradient-descent-code.rl-dynamics-gradient-sketches-toy-adam5e5
Gradient sketches of a toy RL run (Adam, lr 5e-5, rep 3)
Random-projection sketches of per-rollout policy gradients from a small GRPO run in Apollo Research's RL-dynamics
project (base model Qwen/Qwen3.5-4B with LoRA adapters, 129,859,584 trainable parameters, 20 training steps,
1,024 rollouts per step). Every gradient was projected with the same subsampled randomised Hadamard transform (SRHT)
to k = 262,144 coordinates, so inner products between sketches approximate inner… See the full description on the dataset page: https://huggingface.co/datasets/apollo-research/rl-dynamics-gradient-sketches-toy-adam5e5.repro-on-the-theory-of-continual-learning-with-gradient-descent-for-neural-networks-traces
Agent traces
Agent sessions published from a Trackio Logbook.
repro-on-the-theory-of-continual-learning-with-gradient-descent-for-neural-networks-traces
Agent traces
Agent sessions published from a Trackio Logbook.
details_gradientai__Llama-3-70B-Instruct-Gradient-524krepro-a-sketch-and-project-analysis-of-subsampled-natural-gradient-algorithms-traces
Agent traces
Agent sessions published from a Trackio Logbook.
env_training_gradientsGradient-Decomposition-Assay
Gradient Decomposition Assay (GDA)
This repository contains the CSV corpus and summary tables for the Gradient Decomposition Assay, an exploratory behavioral evaluation of how frontier language models respond to an eight-vector prompt manifold ranging from benign technical tasks to adversarial compression and counterfactual/narrative reframing.
Why this repository uses multiple configurations
The CSV files in this dataset are not all the same table. The row-level… See the full description on the dataset page: https://huggingface.co/datasets/devinendorphin/Gradient-Decomposition-Assay.repro-chain-of-thought-gradient-descent-code-sol
Chain-of-Thought Gradient Descent — independent reproduction
This directory contains a scaled, self-contained reproduction of the mechanisms
behind ICML 2026 paper #443, Chain-of-Thought Gradient Descent.
The paper does not provide code or complete training hyperparameters. This
reproduction therefore tests its two central mechanisms directly:
A frozen, one-layer decoder-style attention block is trained once and reused
to emit local ReLU-network forward and backward blocks. It… See the full description on the dataset page: https://huggingface.co/datasets/JG1310/repro-chain-of-thought-gradient-descent-code-sol.details_gradientai__Llama-3-8B-Instruct-262kdetails_gradientai__Llama-3-8B-Instruct-Gradient-1048kGradient-Reasoning