fix(factory.py): correctly apply bos token for llama3 instruct prompt…

… template
azohra · Apr 26, 2024 · ccb864b · ccb864b
1 parent 5bd50a6
commit ccb864b
Showing 1 changed file with 2 additions and 8 deletions.
diff --git a/litellm/llms/prompt_templates/factory.py b/litellm/llms/prompt_templates/factory.py
@@ -1350,11 +1350,8 @@ def prompt_factory(
  return llama_2_chat_pt(messages=messages)
  elif "llama3" in model and "instruct" in model:
  return hf_chat_template(
- model=model,
+ model="meta-llama/Meta-Llama-3-8B-Instruct",
  messages=messages,
- chat_template=known_tokenizer_config[ # type: ignore
- "meta-llama/Meta-Llama-3-8B-Instruct"
- ]["tokenizer"]["chat_template"],
  )
  elif custom_llm_provider == "perplexity":
  for message in messages:
@@ -1369,11 +1366,8 @@ def prompt_factory(
  "meta-llama/llama-3" in model or "meta-llama-3" in model
  ) and "instruct" in model:
  return hf_chat_template(
- model=model,
+ model="meta-llama/Meta-Llama-3-8B-Instruct",
  messages=messages,
- chat_template=known_tokenizer_config[ # type: ignore
- "meta-llama/Meta-Llama-3-8B-Instruct"
- ]["tokenizer"]["chat_template"],
  )
  elif (
  "tiiuae/falcon" in model