ContraCLM: Contrastive Learning For Causal Language Model
This work addresses the limitation of causal language models for tasks beyond language generation, such as semantic similarity and code search, though it appears incremental in enhancing existing models.
The paper tackles the problem of poor discrimination ability in causal language models by introducing ContraCLM, a contrastive learning framework that improves representation expressiveness, resulting in a 44% relative improvement on Semantic Textual Similarity tasks and a 9% relative improvement on execution accuracy for source code generation.
Despite exciting progress in causal language models, the expressiveness of the representations is largely limited due to poor discrimination ability. To remedy this issue, we present ContraCLM, a novel contrastive learning framework at both token-level and sequence-level. We assess ContraCLM on a variety of downstream tasks. We show that ContraCLM enhances discrimination of the representations and bridges the gap with the encoder-only models, which makes causal language models better suited for tasks beyond language generation. Specifically, we attain $44\%$ relative improvement on the Semantic Textual Similarity tasks and $34\%$ on Code-to-Code Search tasks. Furthermore, by improving the expressiveness of the representations, ContraCLM also boosts the source code generation capability with $9\%$ relative improvement on execution accuracy on the HumanEval benchmark.