This model is randomly initialized, using the config from Qwen/Qwen1.5-MoE-A2.7B-Chat but with smaller size. Note the model is in float16.

Codes:

import transformers
import torch
import os
from huggingface_hub import create_repo, upload_folder

source_model_id = 'Qwen/Qwen1.5-MoE-A2.7B-Chat'
save_path = '/tmp/yujiepan/qwen1.5-moe-tiny-random'
repo_id = 'yujiepan/qwen1.5-moe-tiny-random'

config = transformers.AutoConfig.from_pretrained(
    source_model_id, trust_remote_code=True)
config.hidden_size = 4
config.intermediate_size = 2
config.num_attention_heads = 4
config.num_hidden_layers = 2
config.num_key_value_heads = 2
config.moe_intermediate_size = 2
config.shared_expert_intermediate_size = 2
config.max_window_layers = 1
config.use_sliding_window = True
config.torch_dtype = torch.float16

model = transformers.AutoModelForCausalLM.from_config(
    config, trust_remote_code=True, torch_dtype=torch.float16)
model = model.half()

tokenizer = transformers.AutoTokenizer.from_pretrained(
    source_model_id, trust_remote_code=True)

result = transformers.pipelines.pipeline(
    'text-generation',
    model=model, tokenizer=tokenizer,
    device=0,
    max_new_tokens=16,
)('Hello World!')
print(result)

model.save_pretrained(save_path)
tokenizer.save_pretrained(save_path)

os.system(f'ls -alh {save_path}')
create_repo(repo_id, exist_ok=True)
upload_folder(repo_id=repo_id, folder_path=save_path)
Downloads last month
12
Safetensors
Model size
1.22M params
Tensor type
FP16
·
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Collection including yujiepan/qwen1.5-moe-tiny-random