TMaxxx/tmax-27b-v1.1
Reinforcement Learning • 2.65M • Updated • 79
Datasets:
Tasks:
Text Generation
Modalities:
Text
Formats:
parquet
Languages:
English
Size:
10K - 100K
ArXiv:
License:
messages listlengths 2 2 | ground_truth stringlengths 20 20 | dataset stringclasses 1
value | env_config dict | source stringclasses 1
value |
|---|---|---|---|---|
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000000_c19dda5b | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:618e344e0172",
"task_id": "task_000000_c19dda5b"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000000_f8baca82 | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishivi/swerl-tmax-hardened:task_000000_f8baca82-0c4e432c",
"task_id": "task_000000_f8baca82"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000001_1a03ebc5 | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:37a79d0fd9b9",
"task_id": "task_000001_1a03ebc5"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000001_e67d5628 | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:f4f76c1a90bf",
"task_id": "task_000001_e67d5628"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000002_81d660f3 | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:ac0931df29ee",
"task_id": "task_000002_81d660f3"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000002_8be5378a | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishivi/swerl-tmax-hardened:task_000002_8be5378a-0c4e432c",
"task_id": "task_000002_8be5378a"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000003_2a1af405 | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishivi/swerl-tmax-hardened:task_000003_2a1af405-0c4e432c",
"task_id": "task_000003_2a1af405"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000003_b73a4413 | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:d5886f5c569e",
"task_id": "task_000003_b73a4413"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000004_aeddda76 | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:7da96e1618b7",
"task_id": "task_000004_aeddda76"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000004_b4949f3f | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:d5fc18e4404f",
"task_id": "task_000004_b4949f3f"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000005_ceefb305 | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:14b0edbb09bf",
"task_id": "task_000005_ceefb305"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000006_74a36316 | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:98957e1d21c4",
"task_id": "task_000006_74a36316"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000006_93ea43e6 | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:34eff4a10087",
"task_id": "task_000006_93ea43e6"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000007_654ab1c7 | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:27efbe8735b4",
"task_id": "task_000007_654ab1c7"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000007_eadb8636 | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishivi/swerl-tmax-hardened:task_000007_eadb8636-0c4e432c",
"task_id": "task_000007_eadb8636"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000008_e94be48f | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:be77886e06e4",
"task_id": "task_000008_e94be48f"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000009_dbc34e08 | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:6dc3741227cc",
"task_id": "task_000009_dbc34e08"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000010_7c427a6f | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:363b894bc5df",
"task_id": "task_000010_7c427a6f"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000010_e7b016b0 | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:ff1557b69235",
"task_id": "task_000010_e7b016b0"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000011_24e54d45 | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:2df1439e90a7",
"task_id": "task_000011_24e54d45"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000011_b3a63e5b | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:8d9341b5b4ea",
"task_id": "task_000011_b3a63e5b"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000012_58fdcccc | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:9ba2dcb5f85d",
"task_id": "task_000012_58fdcccc"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000012_bcc9081e | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:87d74314a166",
"task_id": "task_000012_bcc9081e"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000013_df6cd2ad | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:0ba14e3e653c",
"task_id": "task_000013_df6cd2ad"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000014_3d9789ba | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:d298403af06c",
"task_id": "task_000014_3d9789ba"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000014_a15a2b8a | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:d37290e1435a",
"task_id": "task_000014_a15a2b8a"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000015_6615e8e8 | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishivi/swerl-tmax-hardened:task_000015_6615e8e8-0c4e432c",
"task_id": "task_000015_6615e8e8"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000015_802ac83d | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:ac4eacb9ab8c",
"task_id": "task_000015_802ac83d"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000016_3ffaf505 | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:40e08ab795a8",
"task_id": "task_000016_3ffaf505"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000017_0ba5f186 | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:491f76904c02",
"task_id": "task_000017_0ba5f186"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000017_75ba7ed9 | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishivi/swerl-tmax-hardened:task_000017_75ba7ed9-0c4e432c",
"task_id": "task_000017_75ba7ed9"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000018_3997ce32 | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:4f58e58929c0",
"task_id": "task_000018_3997ce32"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000018_42ce2436 | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:6dbf396d3ce7",
"task_id": "task_000018_42ce2436"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000019_9b7bd652 | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishivi/swerl-tmax-hardened:task_000019_9b7bd652-0c4e432c",
"task_id": "task_000019_9b7bd652"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000019_d40ae014 | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:a7345ae4122b",
"task_id": "task_000019_d40ae014"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000020_3c0b25ea | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishivi/swerl-tmax-hardened:task_000020_3c0b25ea-0c4e432c",
"task_id": "task_000020_3c0b25ea"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000020_bdaab293 | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:9e859d6da0f0",
"task_id": "task_000020_bdaab293"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000021_5e1d9408 | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:3c5925451d0d",
"task_id": "task_000021_5e1d9408"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000021_ab956b8c | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:c0919314e549",
"task_id": "task_000021_ab956b8c"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000022_21389603 | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:5e21650edf06",
"task_id": "task_000022_21389603"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000023_004a85b4 | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:e43eb422cf80",
"task_id": "task_000023_004a85b4"
} | tmax_skill_taxonomy |
[
{
"content": "You are a helpful coding assistant. You have access to a bash terminal. Use it to explore the codebase, understand the problem, implement a solution, and verify it works. When you are confident your solution is correct, submit by running: echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT",
"role": "s... | task_000023_97c7568e | passthrough | {
"env_name": "swerl_vanillux_sandbox",
"image": "hamishi740/swerl-tmax-v3:33db6be16bbd",
"task_id": "task_000023_97c7568e"
} | tmax_skill_taxonomy |
End of preview. Expand in Data Studio
💻 Code · 🤗 Models & Data · 📜 Paper · 📓 Blog
TMax 13K v1.1
This is the 13,565-task RL dataset for the TMax v1.1 models, formatted for our Open Instruct fork. It uses the same five-column training format as the original TMax Open Instruct release.
TMax v1.1 hardens and prefilters the original 14,601-task TMax corpus, including verifier repairs and environment fixes. It includes tasks without a demonstrated passing solution. The paper describes the original data generation recipe; the audit notes describe the v1.1 hardening and validation results.
For the task details and audit metadata, see TMaxxx/tmax-13k-v1.1-metadata.
Harbor format: TMaxxx/tmax-13k-v1.1-harbor.
Usage
from datasets import load_dataset
from huggingface_hub import hf_hub_download
ds = load_dataset("TMaxxx/tmax-13k-v1.1", split="train")
archive = hf_hub_download(
"TMaxxx/tmax-13k-v1.1", "task-data.tar.gz", repo_type="dataset",
)
Dataset contents
| Column | Contents |
|---|---|
messages |
System and user messages, as a list of {content, role} objects |
ground_truth |
Task ID used to select the verifier |
dataset |
passthrough |
env_config |
{env_name, image, task_id}; environment is swerl_vanillux_sandbox |
source |
tmax_skill_taxonomy |
All image references are public Docker Hub tags for Linux/AMD64. The published conversion retains 11,852 original images and supplies 1,713 hardened images. Image digests and repair provenance are recorded in image-manifest.jsonl and image-publication-plan.json.
The archive contains the hardened instructions, setup scripts, verifiers, and supporting assets. Use it as the task-data archive in the TMax Open Instruct fork; loading the Parquet alone does not install the task environments.
Both variants contain the same 13,565 task IDs and the same task archive (SHA-256: a654638ec448403823273b6510eeb457ae6d9914c7bf4f87ce5302d1fb0c7146). dataset_release.json pins the source revisions and records the publication checks. This packaging does not rerun task solutions or establish that every task is solvable. Known verifier defects and validation limitations remain documented in AUDIT.md.
License
This release retains the other license marker from the hardened source. The original TMax dataset is licensed under ODC-BY. Consult both sources for reuse terms.
The original data includes Gemini 3.1 Pro outputs subject to Google's terms. The original release also refers to Ai2's Responsible Use Guidelines.
Citation
If you use TMax models or data, please cite:
@misc{ivison2026tmaxsimplerecipeterminal,
title={Tmax: A simple recipe for terminal agents},
author={Hamish Ivison and Junjie Oscar Yin and Rulin Shao and Teng Xiao and Nathan Lambert and Hannaneh Hajishirzi},
year={2026},
eprint={2606.23321},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2606.23321},
}
- Downloads last month
- 2,624
