apunkt/doc-to-lora
1
0
Fork 0
mirror of https://github.com/SakanaAI/doc-to-lora.git synced 2026-07-26 17:11:02 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

hf data

Browse source
This commit is contained in:
51616 2025-06-26 16:23:14 +09:00
parent f08831d957
commit 4cedfc2514
2 changed files with 41 additions and 31 deletions
Download patch file Download diff file Expand all files Collapse all files

43
data/self_generate_qa.py
View file

@ -13,9 +13,10 @@ from ctx_to_lora.data.processing import load_and_process_dataset
from ctx_to_lora.utils import clear_gpu
SYSTEM_TEMPLATE = (
"### System Instructions ###\n"
"### SYSTEM INSTRUCTIONS ###\n"
"You are a creative and helpful assistant.\n"
"**DO NOT** hallucinate or make up information."
"**DO NOT** hallucinate or make up information.\n"
"### END OF SYSTEM INSTRUCTIONS ###"
)
PROMPT_TEMPLATE = "### Context ###\n{context}\n\n\n### Question ###\n{question}"
@ -81,7 +82,7 @@ def get_dataset_configs(
def create_messages(
ctxs: list[str], questions: list[str], vllm_model: str, system_template: str
ctxs: list[str], questions: list[list[str]], vllm_model: str, system_template: str
) -> list[list[dict]]:
"""Create chat messages for the model."""
if "gemma" in vllm_model:
@ -93,7 +94,8 @@ def create_messages(
"content": system_template + "\n\n\n" + get_prompt(ctx, q),
}
]
for ctx, q in zip(ctxs, questions)
for ctx, q_list in zip(ctxs, questions)
for q in q_list
]
else:
return [
@ -101,7 +103,8 @@ def create_messages(
{"role": "system", "content": system_template},
{"role": "user", "content": get_prompt(ctx, q)},
]
for ctx, q in zip(ctxs, questions)
for ctx, q_list in zip(ctxs, questions)
for q in q_list
]
@ -139,18 +142,16 @@ def self_generate(
ds = ds.take(10)
ctxs = [sample["context"] for sample in ds]
questions = [sample["prompt"] for sample in ds]
questions = [sample["prompts"] for sample in ds]
messages = create_messages(ctxs, questions, args.vllm_model, system_template)
messages = create_messages(ctxs, questions, args.vllm_model, SYSTEM_TEMPLATE)
print(f"Generating from {len(messages)} contexts")
completions = llm.chat(
messages,
sampling_params=SamplingParams(
max_tokens=2048,
temperature=0.1
if args.vllm_model == "mistralai/Mistral-Small-3.1-24B-Instruct-2503"
else 1.0,
temperature=1.0,
),
)
@ -243,24 +244,6 @@ if __name__ == "__main__":
max_model_len=MODEL_CTX_LEN.get(vllm_model),
)
system_template = SYSTEM_TEMPLATE
if vllm_model == "mistralai/Mistral-Small-3.1-24B-Instruct-2503":
mm_kwargs = {"image": 0}
llm_kwargs.update(
{
"tokenizer_mode": "mistral",
"config_format": "mistral",
"load_format": "mistral",
"max_model_len": 2**14,
"limit_mm_per_prompt": mm_kwargs,
}
)
system_template = (
"You are Mistral Small 3.1, a Large Language Model (LLM) created by Mistral AI, a French startup headquartered in Paris.\n"
"You power an AI assistant called Le Chat.\n"
"Your knowledge base was last updated on 2023-10-01.\n\n\n"
) + SYSTEM_TEMPLATE
print(f"{llm_kwargs=}")
llm = LLM(**llm_kwargs)
@ -276,7 +259,7 @@ if __name__ == "__main__":
# Process each dataset
for ds_name, split in dataset_configs:
print(f"Processing dataset: {ds_name}, split: {split}")
self_generate(ds_name, split, args, llm, system_template)
self_generate(ds_name, split, args, llm, SYSTEM_TEMPLATE)
else:
assert args.glob_pattern, (
"glob_pattern must be provided if no ds_names or config"
@ -290,5 +273,5 @@ if __name__ == "__main__":
split=args.split,
args=args,
llm=llm,
system_template=system_template,
system_template=SYSTEM_TEMPLATE,
)