Dataset Viewer
The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.
mattermost-mcp benchmark archive
A fixed archive.db for benchmarking mattermost-mcp:
the same SQLite schema a real sync writes, filled with public chat text and real
embeddings, so that performance and search-quality numbers in pull requests can
be rerun by anyone.
- Source text: "Software-related Slack Chats with Disentangled Conversations" (P. Chatterjee, K. Damevski, N. Kraft, L. Pollock, MSR 2020, Zenodo 3627124): 437,893 messages from the public pythondev, clojurians and elmlang Slack communities, June 2017 to June 2019, with pseudonymous user names.
- Layout: built by
scripts/bench-archive.mjs build(seedmattermost-mcp-bench-v1). Each disentangled conversation keeps its text and is placed, by a hash of its key, into a DM with the archive owner, a group chat, a private channel or one of several public channels, threaded or flat; blank file-only and bot posts are added. Result: 475,298 posts in 878 channels (DMs 36%, group 3%, public 39%, private 22% of posts), 27% thread replies, 8% blank, matching the shape of a real archive. - Embeddings: every non-blank post embedded with the vendored multilingual-e5-small (fp16 on CUDA) and quantized with sqlite-vector 1.1.0 (INT8), exactly as the server's backfill does.
Usage: node scripts/bench-archive.mjs fetch downloads this file at the
revision pinned in the repository and checks its sha256;
node scripts/bench.mjs bench-archive.db runs the benchmark against it.
- Downloads last month
- 59