yujiepan/jamba-1.5-tiny-random

This model is for debugging. It is randomly initialized using the config from ai21labs/AI21-Jamba-1.5-Large but with smaller size.

Codes:

import os

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

model_id = 'ai21labs/AI21-Jamba-1.5-Large'
save_path = '/tmp/yujiepan/jamba-1.5-tiny-random'
repo_id = 'yujiepan/jamba-1.5-tiny-random'

config = transformers.AutoConfig.from_pretrained(
    model_id, trust_remote_code=True)
config.hidden_size = 8
config.intermediate_size = 16
config.num_attention_heads = 4
config.num_hidden_layers = 16
config.num_key_value_heads = 2
# config.use_mamba_kernels = False

model = AutoModelForCausalLM.from_config(
    config, torch_dtype=torch.bfloat16, attn_implementation="sdpa", trust_remote_code=True
)

set_seed(42)
with torch.no_grad():
    for _, p in sorted(model.named_parameters()):
        torch.nn.init.uniform_(p, -0.2, 0.2)

model.generation_config = GenerationConfig.from_pretrained(
    model_id, trust_remote_code=True)
model.save_pretrained(save_path)

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

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