MiniCPM5-1B-oQ4 / config.json
AtBeginDocument's picture
Upload MiniCPM5-1B-oQ4 via oMLX
d0ccb49 verified
Raw
History Blame Contribute Delete
10.9 kB
{
"_name_or_path": "openbmb/MiniCPM5-1B",
"architectures": [
"LlamaForCausalLM"
],
"bos_token_id": 0,
"eos_token_id": [
1,
130073
],
"pad_token_id": 1,
"hidden_act": "silu",
"hidden_size": 1536,
"initializer_range": 0.02,
"intermediate_size": 4608,
"max_position_embeddings": 131072,
"model_type": "llama",
"num_attention_heads": 16,
"num_hidden_layers": 24,
"num_key_value_heads": 2,
"head_dim": 128,
"rms_norm_eps": 1e-06,
"rope_theta": 5000000,
"rope_scaling": null,
"tie_word_embeddings": false,
"torch_dtype": "bfloat16",
"transformers_version": "5.6.2",
"use_cache": true,
"vocab_size": 130560,
"quantization": {
"group_size": 64,
"bits": 4,
"mode": "affine",
"model.layers.0.mlp.down_proj": {
"bits": 6,
"group_size": 64,
"mode": "affine"
},
"model.layers.0.self_attn.k_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.0.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.0.self_attn.q_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.0.self_attn.v_proj": {
"bits": 6,
"group_size": 64,
"mode": "affine"
},
"model.layers.1.mlp.down_proj": {
"bits": 6,
"group_size": 64,
"mode": "affine"
},
"model.layers.1.self_attn.k_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.1.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.1.self_attn.q_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.1.self_attn.v_proj": {
"bits": 6,
"group_size": 64,
"mode": "affine"
},
"model.layers.10.mlp.down_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.10.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.11.mlp.down_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.11.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.12.mlp.down_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.12.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.13.mlp.down_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.13.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.14.mlp.down_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.14.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.15.mlp.down_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.15.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.16.mlp.down_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.16.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.17.mlp.down_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.17.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.18.mlp.down_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.18.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.19.mlp.down_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.19.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.2.mlp.down_proj": {
"bits": 6,
"group_size": 64,
"mode": "affine"
},
"model.layers.2.self_attn.k_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.2.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.2.self_attn.q_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.2.self_attn.v_proj": {
"bits": 6,
"group_size": 64,
"mode": "affine"
},
"model.layers.20.mlp.down_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.20.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.21.mlp.down_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.21.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.22.mlp.down_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.22.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.23.mlp.down_proj": {
"bits": 6,
"group_size": 64,
"mode": "affine"
},
"model.layers.23.self_attn.k_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.23.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.3.self_attn.k_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
}
},
"quantization_config": {
"group_size": 64,
"bits": 4,
"mode": "affine",
"model.layers.0.mlp.down_proj": {
"bits": 6,
"group_size": 64,
"mode": "affine"
},
"model.layers.0.self_attn.k_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.0.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.0.self_attn.q_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.0.self_attn.v_proj": {
"bits": 6,
"group_size": 64,
"mode": "affine"
},
"model.layers.1.mlp.down_proj": {
"bits": 6,
"group_size": 64,
"mode": "affine"
},
"model.layers.1.self_attn.k_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.1.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.1.self_attn.q_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.1.self_attn.v_proj": {
"bits": 6,
"group_size": 64,
"mode": "affine"
},
"model.layers.10.mlp.down_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.10.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.11.mlp.down_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.11.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.12.mlp.down_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.12.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.13.mlp.down_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.13.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.14.mlp.down_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.14.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.15.mlp.down_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.15.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.16.mlp.down_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.16.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.17.mlp.down_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.17.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.18.mlp.down_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.18.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.19.mlp.down_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.19.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.2.mlp.down_proj": {
"bits": 6,
"group_size": 64,
"mode": "affine"
},
"model.layers.2.self_attn.k_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.2.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.2.self_attn.q_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.2.self_attn.v_proj": {
"bits": 6,
"group_size": 64,
"mode": "affine"
},
"model.layers.20.mlp.down_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.20.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.21.mlp.down_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.21.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.22.mlp.down_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.22.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.23.mlp.down_proj": {
"bits": 6,
"group_size": 64,
"mode": "affine"
},
"model.layers.23.self_attn.k_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.23.self_attn.o_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
},
"model.layers.3.self_attn.k_proj": {
"bits": 5,
"group_size": 64,
"mode": "affine"
}
}
}