CoolFace
Datasetpublic

Nutanix/transformers_rag_no_feature_llama70b_llama8b_results

Results for RAG without LLM generated features transformers code using queries provided by HuggingFace team. Description The RAG responses are generated using the dataset: https://huggingface.co/datasets/Nutanix/transformers_processed_data_llama70b. Llama-3-128k-70b-Instruct-awq was used for feature extraction of transformers code(v4.42.3) to create the dataset, and Llama-3-8b-Instruct was used for generation of both RAG and Raw outputs. Data Fields id - The id… See the full description on the dataset page: https://huggingface.co/datasets/Nutanix/transformers_rag_no_feature_llama70b_llama8b_results.

sourceHugging Faceupdated 2y agoView on Hugging Face
1likes15downloads
Dataset Card

Results for RAG without LLM generated features transformers code using queries provided by HuggingFace team.

Description

The RAG responses are generated using the dataset: https://huggingface.co/datasets/Nutanix/transformersprocesseddata_llama70b. Llama-3-128k-70b-Instruct-awq was used for feature extraction of transformers code(v4.42.3) to create the dataset, and Llama-3-8b-Instruct was used for generation of both RAG and Raw outputs.

Data Fields

  1. 1.id - The id for the question/response.
  2. 2.question - The queston used for generating response from the LLM.
  3. 3.generation - The response generated using appropriate context from retrieved files
  4. 4.retrieved_files - The filenames retrieved for the query using BM25
  5. 5.retrieved_files_in_context - The retrieved files that fit in the context of the generation LLM.
  6. 6.retrieval_time - Time taken to retrieve the context(files)
  7. 7.generation_time - Time taken to generate the response from the LLM
  8. 8.completion_tokens - Number of tokens in the generated response
  9. 9.prompt tokens - Number of tokens in the prompt
  10. 10.total tokens - Total number of tokens in prompt + generation