from transformers import pipeline
model_id = "tiny-random/qwen2.5"
pipe = pipeline(
    "text-generation", model=model_id,
    trust_remote_code=True, max_new_tokens=8,
)
print(pipe("Hello World!"))

from transformers import AutoModelForCausalLM, AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    dtype="auto",
    device_map="auto"
)
prompt = "Give me a short introduction to large language model."
messages = [
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=32
)
output_ids = generated_ids[0].tolist()
content = tokenizer.decode(output_ids, skip_special_tokens=False)
print(content)

Codes to create this repo:

Click to expand

python

import json
from pathlib import Path

import torch
from huggingface_hub import hf_hub_download
from transformers import (
    AutoConfig,
    AutoModelForCausalLM,
    AutoTokenizer,
    GenerationConfig,
    pipeline,
    set_seed,
)

source_model_id = "Qwen/Qwen2.5-72B-Instruct"
save_folder = "/tmp/tiny-random/qwen25"

tokenizer = AutoTokenizer.from_pretrained(
    source_model_id, trust_remote_code=True,
)
tokenizer.save_pretrained(save_folder)

with open(hf_hub_download(source_model_id, filename='config.json', repo_type='model'), 'r', encoding='utf-8') as f:
    config_json: dict = json.load(f)
config_json.update({
    "num_hidden_layers": 4,
    "hidden_size": 8,
    "intermediate_size": 32,
    "max_window_layers": 2,
    "head_dim": 32,
    "num_attention_heads": 8,
    "num_key_value_heads": 4,
})
with open(f"{save_folder}/config.json", "w", encoding='utf-8') as f:
    json.dump(config_json, f, indent=2)

config = AutoConfig.from_pretrained(
    save_folder,
    trust_remote_code=True,
)
model = AutoModelForCausalLM.from_config(
    config,
    torch_dtype=torch.bfloat16,
    trust_remote_code=True,
)
model.generation_config = GenerationConfig.from_pretrained(
    source_model_id, trust_remote_code=True,
)
set_seed(42)
with torch.no_grad():
    for name, p in sorted(model.named_parameters()):
        torch.nn.init.normal_(p, 0, 0.2)
        print(name, p.shape)
model.save_pretrained(save_folder)

Printing the model:

Click to expand

text

Qwen2ForCausalLM(
  (model): Qwen2Model(
    (embed_tokens): Embedding(152064, 8)
    (layers): ModuleList(
      (0-3): 4 x Qwen2DecoderLayer(
        (self_attn): Qwen2Attention(
          (q_proj): Linear(in_features=8, out_features=256, bias=True)
          (k_proj): Linear(in_features=8, out_features=128, bias=True)
          (v_proj): Linear(in_features=8, out_features=128, bias=True)
          (o_proj): Linear(in_features=256, out_features=8, bias=False)
        )
        (mlp): Qwen2MLP(
          (gate_proj): Linear(in_features=8, out_features=32, bias=False)
          (up_proj): Linear(in_features=8, out_features=32, bias=False)
          (down_proj): Linear(in_features=32, out_features=8, bias=False)
          (act_fn): SiLUActivation()
        )
        (input_layernorm): Qwen2RMSNorm((8,), eps=1e-06)
        (post_attention_layernorm): Qwen2RMSNorm((8,), eps=1e-06)
      )
    )
    (norm): Qwen2RMSNorm((8,), eps=1e-06)
    (rotary_emb): Qwen2RotaryEmbedding()
  )
  (lm_head): Linear(in_features=8, out_features=152064, bias=False)
)

Test environment:

torch: 2.11.0
transformers: 5.5.0