SWE-chat: Coding Agent Interactions From Real Users in the Wild
Paper • 2604.20779 • Published • 20
Dataset Viewer
The dataset viewer is not available for this subset.
Cannot get the split names for the config 'default' of the dataset.
Exception: SplitsNotFoundError
Message: The split names could not be parsed from the dataset config.
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
for split_generator in builder._split_generators(
~~~~~~~~~~~~~~~~~~~~~~~~~^
StreamingDownloadManager(base_path=builder.base_path, download_config=download_config)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/src/libs/libcommon/src/libcommon/packaged_modules.py", line 18, in _refuse_lance
raise NotImplementedError(LANCE_DISABLED_MESSAGE)
NotImplementedError: The Lance format is not supported.
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/split_names.py", line 68, in compute_split_names_from_streaming_response
for split in get_dataset_split_names(
~~~~~~~~~~~~~~~~~~~~~~~^
path=dataset,
^^^^^^^^^^^^^
config_name=config,
^^^^^^^^^^^^^^^^^^^
token=hf_token,
^^^^^^^^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
info = get_dataset_config_info(
path,
...<6 lines>...
**config_kwargs,
)
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
funes memory
A funes memory: agent sessions chunked, embedded, and written to a Lance table — a derived index holding verbatim passages with exact provenance, not raw transcripts.
Any agent (or you) can recall from it directly — no local index needed:
funes recall "what did we decide about …" --memory huggingface/funes-swe-chat-zh-e5
Get funes:
curl -fsSL https://huggingface.co/buckets/huggingface/funes/resolve/install.sh | sh
Source
527 coding-agent sessions from SALT-NLP/SWE-chat v2
(release commit df393df, ODC-BY): every session of the 38 public GitHub repositories that hold at least one
session whose user prompts are mostly Chinese. 318 of the sessions are mostly Chinese, the others mostly
English, on the same projects. They come from Claude Code (319), OpenCode (142), Cursor (37) and Codex (29),
and hold 4,638 user prompts, 2,584 of them in Chinese.
Each session is converted from the dataset's normalized conversations table into funes's
turns format: a transcript entry becomes
a turn, its rows the turn's blocks (prompts, replies, thinking, tool calls and tool results). Metadata rows are
left out, and so is the one session named unknown, whose id collides across repositories. OpenCode sessions
carry tool calls without their results, as in the source. Embedded with intfloat/multilingual-e5-small, with
an icu text index, by funes 1.9.0. Chunks in which the publish gate found secrets were scrubbed before
publishing (19 of 157,022).
License
ODC-BY, as the source. Please credit SWE-chat: SWE-chat: Real-World AI Coding Sessions in the Wild, Baumann et al., COLM 2026 (arXiv:2604.20779). SWE-chat only includes repositories whose licenses allow redistribution; their code in these sessions stays under those licenses.
| Chunks | 157,003 |
| Embedding model | intfloat/multilingual-e5-small |
| Updated | 2026-10-22 |
Published and kept current by funes push.
- Downloads last month
- 30