datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
SWE-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.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.
pvp-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-on-the-theory-of-continual-learning-with-gradient-descent-for-neural-networks-traces
Agent traces
Agent sessions published from a Trackio Logbook.
repro-gradmem-learning-to-write-context-into-memory-with-test-time-gradient-descent-traces
Agent traces
Agent sessions published from a Trackio Logbook.
dpo-gradients
Gradients DPO comparison
This public copy contains 53,334 training rows and 1,715 test rows. Each row has
the flat string fields prompt, chosen, and rejected expected by Gradients DPO
preparation. It is a lossless field conversion of the local Together-format source:
the user message becomes prompt, the preferred assistant content becomes chosen,
and the non-preferred assistant content becomes rejected.
reasoning-improved-gradients
Gradients reasoning comparison
This public copy contains 100,000 training rows and 1,000 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.
math-gradients
Gradients math comparison
This public copy contains 7,913 training rows and 879 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.
gradientai__Llama-3-8B-Instruct-Gradient-1048k-details
Dataset Card for Evaluation run of gradientai/Llama-3-8B-Instruct-Gradient-1048k
Dataset automatically created during the evaluation run of model gradientai/Llama-3-8B-Instruct-Gradient-1048k
The dataset is composed of 44 configuration(s), each one corresponding 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… See the full description on the dataset page: https://huggingface.co/datasets/open-llm-leaderboard/gradientai__Llama-3-8B-Instruct-Gradient-1048k-details.ultra-feedback-gradientgradientai__Gradient-Llama-3.1-8B-Instruct-1048k-details
Dataset Card for Evaluation run of gradientai/Gradient-Llama-3.1-8B-Instruct-1048k
Dataset automatically created during the evaluation run of model gradientai/Gradient-Llama-3.1-8B-Instruct-1048k
The dataset is composed of 38 configuration(s), each one corresponding 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… See the full description on the dataset page: https://huggingface.co/datasets/open-llm-leaderboard/gradientai__Gradient-Llama-3.1-8B-Instruct-1048k-details.mi-prueba-gradientstest-datasettinyperson-gradient-stability-runs