chore(vllm): codespell for spell checking (#2820)

This commit is contained in:
Massimiliano Pronesti
2024-02-21 18:56:01 -08:00
committed by GitHub
parent 95529e3253
commit 93dc5a2870
16 changed files with 85 additions and 25 deletions
+1 -1
View File
@@ -41,7 +41,7 @@ class DeciLMForCausalLM(LlamaForCausalLM):
Based on the llama executor.
The main difference is that DeciLM uses Variable Grouped Query Attention.
The constant number of GQA heads in the decoder is overriden with a value
The constant number of GQA heads in the decoder is overridden with a value
per layer.
Usually, in the HuggingFace implementation, instead of