Ru-Gemma3-1B

ะžะฟะธัะฐะฝะธะต

ะญั‚ะพ ัะบัะฟะตั€ะธะผะตะฝั‚ะฐะปัŒะฝะฐั ะฒะตั€ัะธั ะผะพะดะตะปะธ Gemma 3 1B, ะดะพะพะฑัƒั‡ะตะฝะฝะฐั ะฝะฐ ั€ัƒััะบะพัะทั‹ั‡ะฝะพะผ ะดะฐั‚ะฐัะตั‚ะต Saiga-scored. ะฆะตะปัŒ ะดะพะพะฑัƒั‡ะตะฝะธั โ€” ะฟะพะดั‚ัะฝัƒั‚ัŒ ะบะฐั‡ะตัั‚ะฒะพ ะพะฑั‰ะตะฝะธั ะฝะฐ ั€ัƒััะบะพะผ ัะทั‹ะบะต ะธ ะฐะดะฐะฟั‚ะธั€ะพะฒะฐั‚ัŒ ะผะพะดะตะปัŒ ะฟะพะด ั„ะพั€ะผะฐั‚ "Assistant/User".

ะ’ะฝะธะผะฐะฝะธะต:

  • ะœะพะดะตะปัŒ ะพะฑัƒั‡ะตะฝะฐ ะฒัะตะณะพ 1 ัะฟะพั…ัƒ (ัะบัะฟะตั€ะธะผะตะฝั‚ะฐะปัŒะฝั‹ะน ั€ะฐะฝ).
  • ะญั‚ะพ 1B ะฟะฐั€ะฐะผะตั‚ั€ะพะฒ (ะพั‡ะตะฝัŒ ะผะฐะปะตะฝัŒะบะฐั ะผะพะดะตะปัŒ), ะฟะพัั‚ะพะผัƒ ะฝะต ะถะดะธั‚ะต ะพั‚ ะฝะตะต ั‡ัƒะดะตั ัƒั€ะพะฒะฝั GPT-5.
  • ะ’ ะพั‚ะฒะตั‚ะฐั… ะฒะพะทะผะพะถะฝะฐ "ะบะฐัˆะฐ" (ัะผะตัˆะธะฒะฐะฝะธะต ัะทั‹ะบะพะฒ, ะณะฐะปะปัŽั†ะธะฝะฐั†ะธะธ, ะฟะพั‚ะตั€ั ะบะพะฝั‚ะตะบัั‚ะฐ).
  • ะ˜ัะฟะพะปัŒะทัƒะนั‚ะต ะฝะฐ ัะฒะพะน ัั‚ั€ะฐั… ะธ ั€ะธัะบ. ะะฒั‚ะพั€ ะฝะต ะฝะตัะตั‚ ะพั‚ะฒะตั‚ัั‚ะฒะตะฝะฝะพัั‚ะธ ะทะฐ ัะณะตะฝะตั€ะธั€ะพะฒะฐะฝะฝั‹ะน ะบะพะฝั‚ะตะฝั‚.

ะ”ะตั‚ะฐะปะธ ะพะฑัƒั‡ะตะฝะธั

  • Base Model: Gemma 3 1B Instruct
  • Dataset: Saiga Scored (~40k ะดะธะฐะปะพะณะพะฒ)
  • Hardware: NVIDIA RTX 4070
  • Library: Unsloth (QLoRA)
  • Epochs: 1

ะšะฐะบ ะทะฐะฟัƒัั‚ะธั‚ัŒ

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "wexyyyyyy/Ru-Gemma3-1B"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto"
)

messages = [
    {"role": "user", "content": "ะŸั€ะธะฒะตั‚! ะ ะฐััะบะฐะถะธ, ะฟะพั‡ะตะผัƒ ะฝะตะฑะพ ะณะพะปัƒะฑะพะต?"}
]

input_ids = tokenizer.apply_chat_template(
    messages,
    add_generation_prompt=True,
    return_tensors="pt"
).to(model.device)

outputs = model.generate(
    input_ids,
    max_new_tokens=512,
    temperature=0.6,
    top_p=0.9
)

print(tokenizer.decode(outputs[0][input_ids.shape[-1]:], skip_special_tokens=True))
Downloads last month
16
Safetensors
Model size
1.0B params
Tensor type
BF16
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for wexyyyyyy/Ru-Gemma3-1B

Finetuned
(581)
this model
Finetunes
1 model
Quantizations
1 model

Dataset used to train wexyyyyyy/Ru-Gemma3-1B