metadata
license: mit
language:
- en
pretty_name: PIPPA 174K
PIPKIN 174K Creative
The PIPKIN 174K Creative dataset is the largest dataset in the series of evolving, in-the-wild creative datasets. This dataset is inspired by Pygmalion's PIPPA dataset from 2023.
Data is collected by exchanging anonymous data for OSS model usage (GLM-5, GLM 4.7, DeepSeek V3.1, Qwen 3.5 397B, Kimi K2.5, etc).
This data shows 1.6 billion tokens of chat data. The average amount of input tokens is ~1.2k. The average amount of completion tokens is ~1235. There are 174k examples of unique conversations.
Models featured:
deepseek-ai/deepseek-v3.2
deepseek-ai/deepseek-v3.1
moonshotai/kimi-k2-thinking
moonshotai/kimi-k2.5
z-ai/glm4.7
moonshotai/kimi-k2-instruct-0905
stepfun-ai/step-3.5-flash
mistralai/mistral-large-3-675b-instruct-2512
qwen/qwen3-coder-480b-a35b-instruct
google/gemma-3-27b-it
nvidia/nemotron-3-super-120b-a12b
qwen/qwen3.5-122b-a10b
minimaxai/minimax-m2.5
qwen/qwen3.5-397b-a17b
z-ai/glm5
openai/gpt-oss-120b
openai/gpt-oss-20b
qwen/qwq-32b
mistralai/devstral-2-123b-instruct-2512
qwen/qwen3-next-80b-a3b-instruct
qwen/qwen3-next-80b-a3b-thinking
bytedance/seed-oss-36b-instruct
google/gemma-3n-e4b-it
google/gemma-3n-e2b-it
google/gemma-4-31b-it
pipipipipi