You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

Bộ đo tốc độ: 1000 tính năng + 1000 agent

Request chat/completions để đo tốc độ model qua LLM Gateway MISA (30/09/2026). Nguồn: log gateway thật ngày 04/09/2026 (66.680 request, 10 dự án). Dữ liệu khách hàng thật, không công khai.

config mẫu nguồn
tinh_nang 1000 log thật, gọi LLM một lượt, chia theo nhóm system prompt và tỉ lệ lưu lượng
agent 1000 206 log thật (có tools hoặc nhiều lượt) + 794 sinh bằng Petri

Phần agent sinh ra: system prompt THẬT + tools THẬT của 13 loại agent, auditor GLM-5.3-Flash đóng vai người dùng và trả kết quả tool, target misa-ai-2.0; mỗi lượt gọi target là một request.

Cột chính:

  • request: chuỗi JSON của request OpenAI gốc (json.loads rồi gửi thẳng).
  • output_tokens: số token output thật, dùng đặt max_tokens khi đo; output_uoc_luong=true là trung vị nhóm (request stream không lưu response).
  • nguon: that hoặc sinh; app: mã dự án; model_that: model trong log.
  • processing_time_ms: thời gian gateway ghi nhận (chỉ có ở mẫu thật).
Downloads last month
5