-
-
Notifications
You must be signed in to change notification settings - Fork 958
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
* add dpo llama3 * fix dpo bos and eos * bos token gets added automatically by the tokenizer * explicit <|end_of_text|> not needed, as eot_id is sufficient --------- Co-authored-by: Nero10578 <owenarliawan@gmail.com>
- Loading branch information
Showing
1 changed file
with
133 additions
and
0 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,133 @@ | ||
""" | ||
DPO strategies for llama-3 chat template | ||
""" | ||
|
||
|
||
def argilla( | ||
cfg, | ||
**kwargs, | ||
): # pylint: disable=possibly-unused-variable,unused-argument | ||
def transform_fn(sample): | ||
if "system" in sample and sample["system"]: | ||
sample["prompt"] = ( | ||
f"<|start_header_id|>system<|end_header_id|>\n\n{sample['system']}<|eot_id|>" | ||
f"<|start_header_id|>user<|end_header_id|>\n\n{sample['instruction']}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n" | ||
) | ||
else: | ||
sample[ | ||
"prompt" | ||
] = f"<|start_header_id|>user<|end_header_id|>\n\n{sample['instruction']}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n" | ||
sample["chosen"] = f"{sample['chosen_response']}<|eot_id|>" | ||
sample["rejected"] = f"{sample['rejected_response']}<|eot_id|>" | ||
return sample | ||
|
||
return transform_fn | ||
|
||
|
||
def argilla_chat( | ||
cfg, | ||
**kwargs, | ||
): # pylint: disable=possibly-unused-variable,unused-argument | ||
""" | ||
for argilla/dpo-mix-7k conversations | ||
""" | ||
|
||
def transform_fn(sample): | ||
sample[ | ||
"prompt" | ||
] = f"<|start_header_id|>user<|end_header_id|>\n\n{sample['chosen'][0]['content']}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n" | ||
sample["chosen"] = f"{sample['chosen'][1]['content']}<|eot_id|>" | ||
sample["rejected"] = f"{sample['rejected'][1]['content']}<|eot_id|>" | ||
return sample | ||
|
||
return transform_fn | ||
|
||
|
||
def icr( | ||
cfg, | ||
**kwargs, | ||
): # pylint: disable=possibly-unused-variable,unused-argument | ||
""" | ||
chatml transforms for datasets with system, input, chosen, rejected | ||
ex. https://huggingface.co/datasets/argilla/distilabel-intel-orca-dpo-pairs | ||
""" | ||
|
||
def transform_fn(sample): | ||
if "system" in sample and sample["system"]: | ||
sample["prompt"] = ( | ||
f"<|start_header_id|>system<|end_header_id|>\n\n{sample['system']}<|eot_id|>" | ||
f"<|start_header_id|>user<|end_header_id|>\n\n{sample['input']}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n" | ||
) | ||
else: | ||
sample[ | ||
"prompt" | ||
] = f"<|start_header_id|>user<|end_header_id|>\n\n{sample['input']}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n" | ||
sample["chosen"] = f"{sample['chosen']}<|eot_id|>" | ||
sample["rejected"] = f"{sample['rejected']}<|eot_id|>" | ||
return sample | ||
|
||
return transform_fn | ||
|
||
|
||
def intel(cfg, **kwargs): # pylint: disable=possibly-unused-variable,unused-argument | ||
""" | ||
For Intel Orca DPO Pairs | ||
""" | ||
|
||
def transform_fn(sample): | ||
if "system" in sample and sample["system"]: | ||
sample["prompt"] = ( | ||
f"<|start_header_id|>system<|end_header_id|>\n\n{sample['system']}<|eot_id|>" | ||
f"<|start_header_id|>user<|end_header_id|>\n\n{sample['question']}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n" | ||
) | ||
else: | ||
sample[ | ||
"prompt" | ||
] = f"<|start_header_id|>user<|end_header_id|>\n\n{sample['question']}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n" | ||
sample["chosen"] = f"{sample['chosen']}<|eot_id|>" | ||
sample["rejected"] = f"{sample['rejected']}<|eot_id|>" | ||
return sample | ||
|
||
return transform_fn | ||
|
||
|
||
def prompt_pairs( | ||
cfg, **kwargs | ||
): # pylint: disable=possibly-unused-variable,unused-argument | ||
def transform_fn(sample): | ||
if "system" in sample and sample["system"]: | ||
sample["prompt"] = ( | ||
f"<|start_header_id|>system<|end_header_id|>\n\n{sample['system']}<|eot_id|>" | ||
f"<|start_header_id|>user<|end_header_id|>\n\n{sample['prompt']}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n" | ||
) | ||
else: | ||
sample[ | ||
"prompt" | ||
] = f"<|start_header_id|>user<|end_header_id|>\n\n{sample['prompt']}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n" | ||
sample["chosen"] = f"{sample['chosen']}<|eot_id|>" | ||
sample["rejected"] = f"{sample['rejected']}<|eot_id|>" | ||
return sample | ||
|
||
return transform_fn | ||
|
||
|
||
def ultra(cfg, **kwargs): # pylint: disable=possibly-unused-variable,unused-argument | ||
""" | ||
for ultrafeedback binarized conversations | ||
""" | ||
|
||
def transform_fn(sample): | ||
if "system" in sample and sample["system"]: | ||
sample["prompt"] = ( | ||
f"<|start_header_id|>system<|end_header_id|>\n\n{sample['system']}<|eot_id|>" | ||
f"<|start_header_id|>user<|end_header_id|>\n\n{sample['prompt']}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n" | ||
) | ||
else: | ||
sample[ | ||
"prompt" | ||
] = f"<|start_header_id|>user<|end_header_id|>\n\n{sample['prompt']}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n" | ||
sample["chosen"] = f"{sample['chosen'][1]['content']}<|eot_id|>" | ||
sample["rejected"] = f"{sample['rejected'][1]['content']}<|eot_id|>" | ||
return sample | ||
|
||
return transform_fn |
2147cf6
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Oh I see so the bos token gets added automatically by the tokenizer and the eos is not needed.