metadata
configs:
- config_name: multi-turn_chat
data_files:
- split: test
path: multi-turn_chat.parquet
- config_name: code_completion
data_files:
- split: test
path: code_completion.parquet
- config_name: instruction_tuning
data_files:
- split: test
path: instruction_tuning.parquet
- config_name: code_fixing
data_files:
- split: test
path: code_fixing.parquet
- config_name: rag
data_files:
- split: test
path: rag.parquet
- config_name: large_summarization
data_files:
- split: test
path: large_summarization.parquet
- config_name: docstring
data_files:
- split: test
path: docstring.parquet
This dataset contains inference performance benchmarking obtained with vllm version 0.6.1.post2 on different use-case scenarios. The scenarios are defined as bellow:
| Use case | Prompt tokens | Generated tokens |
| Code Completion | 256 | 1024 |
| Docstring Generation | 768 | 128 |
| Code Fixing | 1024 | 1024 |
| RAG | 1024 | 128 |
| Instruction Following | 256 | 128 |
| Multi-turn chat | 512 | 256 |
| Large Summarization | 4096 | 512 |
Benchmarking was conducted with GuideLLM using the following syntax:
guidellm --model <model name> --data-type emulated --data "prompt_tokens=<prompt tokens>,generated_tokens=<generated tokens>"