Skip to content

GitLab

Commit 3e27f9dd authored by daniel-furman's avatar daniel-furman
Browse files

first stab at wrap_chat_template, formatting in function

parent 53c68db4
Hide whitespace changes
Inline Side-by-side
Showing with 3 additions and 1 deletion
lm_eval/models/huggingface.py
View file @ 3e27f9dd
...@@ -664,10 +664,10 @@ class HFLM(LM): ...@@ -664,10 +664,10 @@ class HFLM(LM):
return self.tokenizer.decode(tokens, skip_special_tokens=True) return self.tokenizer.decode(tokens, skip_special_tokens=True)
def tok_wrap_chat_template(self, requests: List[Instance], system: bool = False) -> List[Instance]: def tok_wrap_chat_template(self, requests: List[Instance], system: bool = False) -> List[Instance]:
new_reqs = [] new_reqs = []
for req in requests: for req in requests:
context, continuation = req.args[0].strip(), req.args[1].strip() context, continuation = req.args[0].strip(), req.args[1].strip()
if system: if system:
chat = [ chat = [
{"role": "system", "content": system}, {"role": "system", "content": system},
...@@ -684,11 +684,13 @@ class HFLM(LM): ...@@ -684,11 +684,13 @@ class HFLM(LM):
tokenize=False, tokenize=False,
add_generation_prompt=True, add_generation_prompt=True,
) )
rfind_continuation = single_tokenized_conversation.rfind(continuation) rfind_continuation = single_tokenized_conversation.rfind(continuation)
context = single_tokenized_conversation[:rfind_continuation] context = single_tokenized_conversation[:rfind_continuation]
continuation = single_tokenized_conversation[rfind_continuation:] continuation = single_tokenized_conversation[rfind_continuation:]
req.args = (context, continuation) req.args = (context, continuation)
new_reqs.append(req) new_reqs.append(req)
return new_reqs return new_reqs
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment