Qwen3-30B-A3B-TopK4-Compressed
μ΄ λͺ¨λΈμ Qwen3-30B-A3B Mixture of Experts λͺ¨λΈμ Top-kλ₯Ό 8μμ 4λ‘ κ°μμμΌ μμΆν λ²μ μ λλ€.
λͺ¨λΈ μ 보
- κΈ°λ³Έ λͺ¨λΈ: Qwen3-30B-A3B
- μμΆ λ°©μ: Top-k Reduction (8 β 4)
- μμΆλ₯ : 59.5%
- MMLU μ νλ: 42.9% (7κ° μΉ΄ν κ³ λ¦¬ νκ· )
μμΆ μΈλΆμ¬ν
- μλ³Έ Top-k: 8κ° μ λ¬Έκ° νμ±ν
- μμΆ Top-k: 4κ° μ λ¬Έκ° νμ±ν
- μμΆ ν¨κ³Ό: λͺ¨λΈ ν¬κΈ°μ μΆλ‘ λΉμ©μ ν¬κ² μ€μ΄λ©΄μλ ν©λ¦¬μ μΈ μ±λ₯ μ μ§
μ±λ₯ νκ°
MMLU λ²€μΉλ§ν¬ 7κ° μΉ΄ν κ³ λ¦¬μμ ν μ€νΈ:
- abstract_algebra
- anatomy
- high_school_mathematics
- formal_logic
- professional_medicine
- high_school_macroeconomics
- global_facts
μ¬μ©λ²
from transformers import AutoTokenizer, AutoModelForCausalLM
model_name = "kyne0127/Qwen3-30B-A3B-TopK4-Compressed"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=True)
# μΆλ‘ μμ
input_text = "What is the capital of France?"
inputs = tokenizer(input_text, return_tensors="pt")
outputs = model.generate(**inputs, max_length=100)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)
λΌμ΄μ μ€
Apache 2.0 License
μ μμ
kyne0127
- Downloads last month
- 18