too slow to convert the llm 7b for two hours

[snowyu]

CPU work on 4.8GHz
Memory 128G

pulsar2 llm_build --input_path qwen2.5-7b-intrstruct --output_path qwen2.5-7b-w4a16/ --kv_cache_len 1023 -w s4 --hidden_state_type bf16 --prefill_len 128 --chip AX650
Config(
    model_name='qwen2.5-7b',
    model_type='qwen2',
    num_hidden_layers=28,
    num_attention_heads=28,
    num_key_value_heads=4,
    hidden_size=3584,
    head_dim=0,
    intermediate_size=18944,
    vocab_size=152064,
    rope_theta=1000000.0,
    max_position_embeddings=32768,
    rope_partial_factor=1.0,
    rms_norm_eps=1e-06,
    norm_type='rms_norm',
    hidden_act='silu',
    hidden_act_param=0.03,
    scale_depth=1.4,
    scale_emb=1,
    dim_model_base=256,
    origin_model_type='',
    quant=False,
    quant_sym=False,
    quant_bits=4,
    quant_group_size=128
)
2025-02-11 12:47:06.984 | SUCCESS  | yamain.command.llm_build:llm_build:134 - prepare llm model done!
building llm decode layers   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 28/28 1:59:42
building llm post layer   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 1/1 0:05:03

btw: what's the BSP_MSP_DIR?

[sh-lee2020]

https://github.com/AXERA-TECH/ax650n_bsp_sdk