seoseo99/qwen2-1_5b-sum_lk_gemini
Qwen2-1.5B-Instruct๋ฅผ ํ๊ตญ์ด ์ฌํ/ํ์ฌ ํ๊ธฐ ์์ฝ ์ฉ๋๋ก ๋ฏธ์ธ์กฐ์ ํ 1.5B ํ๋ผ๋ฏธํฐ ๋ชจ๋ธ์
๋๋ค.
1โ3๋ฌธ์ฅ ๊ฐ๊ฒฐ ์์ฝ, ํต์ฌ ํฌ์ธํธ ์ถ์ถ, ์ฌ๋ฌ ํ๊ธฐ ํฉ๋ณธ ์์ฝ์ ์ ํฉํฉ๋๋ค.
ํ์ผ ๊ตฌ์ฑ
config.jsonโ ๋ชจ๋ธ ์ํคํ ์ฒ ์ค์ (hidden size, layer ์ ๋ฑ). ๊ตฌ์กฐ ์ ๋ณด๋ผ ๋ณดํต ์์ ํ์ง ์์ต๋๋ค.generation_config.jsonโgenerate()์ ๊ธฐ๋ณธ๊ฐ(max_new_tokens, temperature, top_p, ๋ฑ)tokenizer.jsonโ Fast ํ ํฌ๋์ด์ ์ ์ฒด ์ ์(vocab/merges/์ ์ฒ๋ฆฌ ํ์ดํ๋ผ์ธ ํฌํจ)tokenizer_config.jsonโ ํ ํฌ๋์ด์ ๋ฉํ(model_max_length, ํน์ํ ํฐ ์ ์ฑ ๋ฑ)special_tokens_map.jsonโeos/pad๋ฑ ํน์ ํ ํฐ ๋งคํmodel-00001-of-00002.safetensors,model-00002-of-00002.safetensorsโ ๋ชจ๋ธ ๊ฐ์ค์น ์ค๋(shard) ํ์ผmodel.safetensors.index.jsonโ ๊ฐ ํ๋ผ๋ฏธํฐ ํ ์๊ฐ ์ด๋ shard์ ์๋์ง ์ธ๋ฑ์ค ๋งต
Introduction (EN)
Qwen2-1.5B-Instruct fine-tuned for Korean travel/event review summarization (1.5B parameters).
Well-suited for 1โ3 sentence concise summaries, key-point extraction, and aggregating multiple reviews.
Files (EN)
config.jsonโ Model architecture settings (hidden size, number of layers, etc.). Structural info; usually not modified.generation_config.jsonโ Default parameters forgenerate()(e.g.,max_new_tokens,temperature,top_p).tokenizer.jsonโ Full definition of the Fast tokenizer (vocab/merges/preprocessing pipeline).tokenizer_config.jsonโ Tokenizer metadata (model_max_length, special-token policies, etc.).special_tokens_map.jsonโ Mapping for special tokens (e.g.,eos,pad).model-00001-of-00002.safetensors,model-00002-of-00002.safetensorsโ Sharded model weights.model.safetensors.index.jsonโ Index mapping that shows which tensors live in which shard.
Quickstart (Transformers)
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch, unicodedata, re
RID = "seoseo99/qwen2-1_5b-sum_lk_gemini"
tok = AutoTokenizer.from_pretrained(RID, use_fast=True, trust_remote_code=True)
if tok.pad_token is None:
tok.pad_token = tok.eos_token
model = AutoModelForCausalLM.from_pretrained(
RID,
torch_dtype=torch.float32, # GPU๋ฉด bfloat16/auto ๊ฐ๋ฅ
low_cpu_mem_usage=True,
trust_remote_code=True,
).eval()
review = "์ฌ๊ธฐ์ ๋ฆฌ๋ทฐ ๋ณธ๋ฌธ์ ๋ฃ์ผ์ธ์"
sys = ("๋ค์ ํ๊ตญ์ด ๋ฆฌ๋ทฐ ๋ณธ๋ฌธ์ 1~3๋ฌธ์ฅ์ผ๋ก ๊ฐ๊ฒฐํ๊ฒ ์์ฝํ์ธ์. "
"๊ณผ์ฅ/๊ด๊ณ ํค ๊ธ์ง, ์ ๋ชฉ/์ง์ญ/๋ ์ง๋ ์ถ๋ ฅํ์ง ๋ง์ธ์.")
body = unicodedata.normalize("NFKC", review).replace("\n", " ")
msgs = [
{"role": "system", "content": sys},
{"role": "user", "content": "ใ๋ฆฌ๋ทฐ ๋ณธ๋ฌธใ\n" + body},
]
ids = tok.apply_chat_template(msgs, add_generation_prompt=True, return_tensors="pt")
out = model.generate(
ids,
max_new_tokens=180,
num_beams=4,
do_sample=False,
no_repeat_ngram_size=4,
repetition_penalty=1.05,
eos_token_id=tok.eos_token_id,
)
text = tok.decode(out[0, ids.shape[-1]:], skip_special_tokens=True)
text = unicodedata.normalize("NFKC", text).replace("\n", " ")
text = re.sub(r"\s+([\.!?])", r"\1", text).strip()
print(text if text.endswith(('.', '!', '?')) else text + '.')
- Downloads last month
- 4
Model tree for seoseo99/qwen2-1_5b-sum_lk_gemini
Base model
Qwen/Qwen2-1.5B-Instruct