Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

From Artificial Needles to Real Haystacks: Improving Retrieval Capabilities in LLMs by Finetuning on Synthetic Data

Zheyang Xiong, Vasilis Papageorgiou, Kangwook Lee, Dimitris Papailiopoulos

TL;DR

It is found that finetuned LLMs' performance on general benchmarks remains almost constant while LLMs finetuned on other baseline long-context augmentation data can encourage hallucination.

Abstract

Recent studies have shown that Large Language Models (LLMs) struggle to accurately retrieve information and maintain reasoning capabilities when processing long-context inputs. To address these limitations, we propose a finetuning approach utilizing a carefully designed synthetic dataset comprising numerical key-value retrieval tasks. Our experiments on models like GPT-3.5 Turbo and Mistral 7B demonstrate that finetuning LLMs on this dataset significantly improves LLMs' information retrieval and reasoning capabilities in longer-context settings. We present an analysis of the finetuned models, illustrating the transfer of skills from synthetic to real task evaluations (e.g., $10.5\%$ improvement on $20$ documents MDQA at position $10$ for GPT-3.5 Turbo). We also find that finetuned LLMs' performance on general benchmarks remains almost constant while LLMs finetuned on other baseline long-context augmentation data can encourage hallucination (e.g., on TriviaQA, Mistral 7B finetuned on our synthetic data cause no performance drop while other baseline data can cause a drop that ranges from $2.33\%$ to $6.19\%$). Our study highlights the potential of finetuning on synthetic data for improving the performance of LLMs on longer-context tasks.

From Artificial Needles to Real Haystacks: Improving Retrieval Capabilities in LLMs by Finetuning on Synthetic Data

TL;DR

It is found that finetuned LLMs' performance on general benchmarks remains almost constant while LLMs finetuned on other baseline long-context augmentation data can encourage hallucination.

Abstract

Recent studies have shown that Large Language Models (LLMs) struggle to accurately retrieve information and maintain reasoning capabilities when processing long-context inputs. To address these limitations, we propose a finetuning approach utilizing a carefully designed synthetic dataset comprising numerical key-value retrieval tasks. Our experiments on models like GPT-3.5 Turbo and Mistral 7B demonstrate that finetuning LLMs on this dataset significantly improves LLMs' information retrieval and reasoning capabilities in longer-context settings. We present an analysis of the finetuned models, illustrating the transfer of skills from synthetic to real task evaluations (e.g., improvement on documents MDQA at position for GPT-3.5 Turbo). We also find that finetuned LLMs' performance on general benchmarks remains almost constant while LLMs finetuned on other baseline long-context augmentation data can encourage hallucination (e.g., on TriviaQA, Mistral 7B finetuned on our synthetic data cause no performance drop while other baseline data can cause a drop that ranges from to ). Our study highlights the potential of finetuning on synthetic data for improving the performance of LLMs on longer-context tasks.
Paper Structure (27 sections, 11 figures, 2 tables)

This paper contains 27 sections, 11 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Long Context LLMs.
  4. Data-centric AI.
  5. LLM Benchmarks and Evals.
  6. Synthetic dataset of retrieval tasks
  7. Simple dictionary key-value retrieval.
  8. Multi-subkey dictionary key-value retrieval.
  9. Prompt with an answer template.
  10. Experiments and results
  11. Stage 1: Finetuning LLMs on synthetic retrieval tasks
  12. Stage 2: Evaluations on long context retrieval and reasoning tasks
  13. Multi-document question answering (MDQA)
  14. Finding 1:
  15. Finding 2:
  16. ...and 12 more sections

Figures (11)

  • Figure 1: An example prompt with desired answer of simple dictionary key-value retrieval task.
  • Figure 2: An example prompt with desired answer of multi-subkey dictionary key-value retrieval task. Here (141, 623, 616) is the gold key. Note that 141 and 623 in the gold key are also subkeys of other keys.
  • Figure 3: The prompt of the simple dictionary key-value retrieval task is provided with an answer template.
  • Figure 4: Token-level loss on the target answer when provided with (right) and without (left) an answer template, where red indicates high and green low loss.
  • Figure 5: Performance of GPT-3.5 Turbo, Mistral 7B and their corresponding finetuned versions on the MDQA task.
  • ...and 6 more figures