yujiepan/falcon-mamba-tiny-random

This model is for debugging. It is randomly initialized using the config from tiiuae/falcon-mamba-7b but with smaller size.

Codes:

import os

import torch

from huggingface_hub import create_repo, upload_folder
from transformers import (
    AutoModelForCausalLM,
    AutoTokenizer,
    GenerationConfig,
    AutoConfig,
    pipeline,
    set_seed,
)

model_id = "tiiuae/falcon-mamba-7b"
repo_id = "yujiepan/falcon-mamba-tiny-random"
save_path = f"/tmp/{repo_id}"
os.system(f'rm -rf {save_path}')

config = AutoConfig.from_pretrained(model_id)
config.use_cache = True
config.num_hidden_layers = 2
config.hidden_size = 8
config.intermediate_size = 16
config.state_size = 8

tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
tokenizer.save_pretrained(save_path)

model = AutoModelForCausalLM.from_config(
    config, torch_dtype=torch.bfloat16,
    trust_remote_code=True,
)
model.generation_config = GenerationConfig.from_pretrained(
    model_id,
    trust_remote_code=True,
)

set_seed(42)
num_params = 0
with torch.no_grad():
    for name, p in sorted(model.named_parameters()):
        print(name, p.shape)
        torch.nn.init.uniform_(p, -0.5, 0.5)
        num_params += p.numel()
print("Total number of parameters:", num_params)
model.save_pretrained(save_path)

pipe = pipeline(
    "text-generation",
    model=save_path,
    device="cpu",
    trust_remote_code=True,
    max_new_tokens=20,
)
print(pipe("Hello World!"))

# create_repo(repo_id, exist_ok=True)
# upload_folder(repo_id=repo_id, folder_path=save_path, repo_type='model')