QwenMoEAriel
QwenMoEAriel is a Mixture of Experts (MoE) made with the following models using LazyMergekit:
🧩 Configuration
base_model : Qwen/Qwen2-1.5B architecture: qwen experts:
- source_model: Qwen/Qwen2-1.5B
positive_prompts:
- "chat"
- "assistant"
- "tell me"
- "explain"
- "I want"
- source_model: Replete-AI/Replete-Coder-Qwen2-1.5b
positive_prompts:
- "code"
- "python"
- "javascript"
- "programming"
- "algorithm" shared_experts:
- source_model: Qwen/Qwen2-1.5B
positive_prompts: # required by Qwen MoE for "hidden" gate mode, otherwise not allowed
- "chat"
(optional, but recommended:)
residual_scale: 0.1 # downweight output from shared expert to prevent overcooking the model
💻 Usage
!pip install -qU transformers bitsandbytes accelerate einops
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(device)
model = AutoModelForCausalLM.from_pretrained(
"femiari/Qwen2-1.5Moe",
torch_dtype=torch.float16,
ignore_mismatched_sizes=True
).to(device)
tokenizer = AutoTokenizer.from_pretrained("femiari/Qwen2-1.5Moe")
prompt = "Give me a short introduction to large language model."
messages = [
{"role": "system", "content": "You are a helpful assistant."},
{"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(device)
generated_ids = model.generate(
model_inputs.input_ids,
max_new_tokens=512
)
generated_ids = [
output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
print(response)