seoseo99/qwen2-1_5b-sum_lk_gemini

Qwen2-1.5B-Instruct๋ฅผ ํ•œ๊ตญ์–ด ์—ฌํ–‰/ํ–‰์‚ฌ ํ›„๊ธฐ ์š”์•ฝ ์šฉ๋„๋กœ ๋ฏธ์„ธ์กฐ์ •ํ•œ 1.5B ํŒŒ๋ผ๋ฏธํ„ฐ ๋ชจ๋ธ์ž…๋‹ˆ๋‹ค.
1โ€“3๋ฌธ์žฅ ๊ฐ„๊ฒฐ ์š”์•ฝ, ํ•ต์‹ฌ ํฌ์ธํŠธ ์ถ”์ถœ, ์—ฌ๋Ÿฌ ํ›„๊ธฐ ํ•ฉ๋ณธ ์š”์•ฝ์— ์ ํ•ฉํ•ฉ๋‹ˆ๋‹ค.

ํŒŒ์ผ ๊ตฌ์„ฑ

  • config.json โ€” ๋ชจ๋ธ ์•„ํ‚คํ…์ฒ˜ ์„ค์ •(hidden size, layer ์ˆ˜ ๋“ฑ). ๊ตฌ์กฐ ์ •๋ณด๋ผ ๋ณดํ†ต ์ˆ˜์ •ํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค.
  • generation_config.json โ€” generate()์˜ ๊ธฐ๋ณธ๊ฐ’(max_new_tokens, temperature, top_p, ๋“ฑ)
  • tokenizer.json โ€” Fast ํ† ํฌ๋‚˜์ด์ € ์ „์ฒด ์ •์˜(vocab/merges/์ „์ฒ˜๋ฆฌ ํŒŒ์ดํ”„๋ผ์ธ ํฌํ•จ)
  • tokenizer_config.json โ€” ํ† ํฌ๋‚˜์ด์ € ๋ฉ”ํƒ€(model_max_length, ํŠน์ˆ˜ํ† ํฐ ์ •์ฑ… ๋“ฑ)
  • special_tokens_map.json โ€” eos/pad ๋“ฑ ํŠน์ˆ˜ ํ† ํฐ ๋งคํ•‘
  • model-00001-of-00002.safetensors, model-00002-of-00002.safetensors โ€” ๋ชจ๋ธ ๊ฐ€์ค‘์น˜ ์ƒค๋“œ(shard) ํŒŒ์ผ
  • model.safetensors.index.json โ€” ๊ฐ ํŒŒ๋ผ๋ฏธํ„ฐ ํ…์„œ๊ฐ€ ์–ด๋А shard์— ์žˆ๋Š”์ง€ ์ธ๋ฑ์Šค ๋งต

Introduction (EN)

Qwen2-1.5B-Instruct fine-tuned for Korean travel/event review summarization (1.5B parameters).
Well-suited for 1โ€“3 sentence concise summaries, key-point extraction, and aggregating multiple reviews.

Files (EN)

  • config.json โ€” Model architecture settings (hidden size, number of layers, etc.). Structural info; usually not modified.
  • generation_config.json โ€” Default parameters for generate() (e.g., max_new_tokens, temperature, top_p).
  • tokenizer.json โ€” Full definition of the Fast tokenizer (vocab/merges/preprocessing pipeline).
  • tokenizer_config.json โ€” Tokenizer metadata (model_max_length, special-token policies, etc.).
  • special_tokens_map.json โ€” Mapping for special tokens (e.g., eos, pad).
  • model-00001-of-00002.safetensors, model-00002-of-00002.safetensors โ€” Sharded model weights.
  • model.safetensors.index.json โ€” 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 ๊ฐ€๋Šฅ
    low_cpu_mem_usage=True,
    trust_remote_code=True,
).eval()

review = "์—ฌ๊ธฐ์— ๋ฆฌ๋ทฐ ๋ณธ๋ฌธ์„ ๋„ฃ์œผ์„ธ์š”"

sys = ("๋‹ค์Œ ํ•œ๊ตญ์–ด ๋ฆฌ๋ทฐ ๋ณธ๋ฌธ์„ 1~3๋ฌธ์žฅ์œผ๋กœ ๊ฐ„๊ฒฐํ•˜๊ฒŒ ์š”์•ฝํ•˜์„ธ์š”. "
       "๊ณผ์žฅ/๊ด‘๊ณ  ํ†ค ๊ธˆ์ง€, ์ œ๋ชฉ/์ง€์—ญ/๋‚ ์งœ๋Š” ์ถœ๋ ฅํ•˜์ง€ ๋งˆ์„ธ์š”.")
body = unicodedata.normalize("NFKC", review).replace("\n", " ")

msgs = [
    {"role": "system", "content": sys},
    {"role": "user",   "content": "ใ€๋ฆฌ๋ทฐ ๋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
Safetensors
Model size
2B params
Tensor type
F32
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for seoseo99/qwen2-1_5b-sum_lk_gemini

Finetuned
(106)
this model