mirror of
https://git.datalinker.icu/vllm-project/vllm.git
synced 2026-05-19 04:27:02 +08:00
[V1][Spec Decode][Ngram] 1.35x gain -> 1.95x gain on InstructCoder with prompt fix (#18971)
This commit is contained in:
parent
6cac54f4d1
commit
135cf55cd1
@ -865,7 +865,15 @@ class InstructCoderDataset(HuggingFaceDataset):
|
|||||||
for item in self.data:
|
for item in self.data:
|
||||||
if len(sampled_requests) >= num_requests:
|
if len(sampled_requests) >= num_requests:
|
||||||
break
|
break
|
||||||
prompt = f"{item['instruction']}:\n{item['input']}"
|
prompt = f"{item['input']}\n\n{item['instruction']} Just output \
|
||||||
|
the code, do not include any explanation."
|
||||||
|
|
||||||
|
# apply template
|
||||||
|
prompt = tokenizer.apply_chat_template(
|
||||||
|
[{"role": "user", "content": prompt}],
|
||||||
|
add_generation_prompt=True,
|
||||||
|
tokenize=False,
|
||||||
|
)
|
||||||
prompt_len = len(tokenizer(prompt).input_ids)
|
prompt_len = len(tokenizer(prompt).input_ids)
|
||||||
sampled_requests.append(
|
sampled_requests.append(
|
||||||
SampleRequest(
|
SampleRequest(
|
||||||
|
|||||||
@ -880,7 +880,19 @@ class InstructCoderDataset(HuggingFaceDataset):
|
|||||||
for item in self.data:
|
for item in self.data:
|
||||||
if len(sampled_requests) >= num_requests:
|
if len(sampled_requests) >= num_requests:
|
||||||
break
|
break
|
||||||
prompt = f"{item['instruction']}:\n{item['input']}"
|
prompt = f"{item['input']}\n\n{item['instruction']} Just output \
|
||||||
|
the code, do not include any explanation."
|
||||||
|
|
||||||
|
# apply template
|
||||||
|
prompt = tokenizer.apply_chat_template(
|
||||||
|
[{
|
||||||
|
"role": "user",
|
||||||
|
"content": prompt
|
||||||
|
}],
|
||||||
|
add_generation_prompt=True,
|
||||||
|
tokenize=False,
|
||||||
|
)
|
||||||
|
|
||||||
prompt_len = len(tokenizer(prompt).input_ids)
|
prompt_len = len(tokenizer(prompt).input_ids)
|
||||||
sampled_requests.append(
|
sampled_requests.append(
|
||||||
SampleRequest(
|
SampleRequest(
|
||||||
|
|||||||
Loading…
x
Reference in New Issue
Block a user