tllm-checkpoint/config.json

{
    "producer": {
        "name": "modelopt",
        "version": "0.25.0"
    },
    "architecture": "Qwen2MoeForCausalLM",
    "dtype": "bfloat16",
    "logits_dtype": "float16",
    "num_hidden_layers": 24,
    "num_attention_heads": 16,
    "num_key_value_heads": 16,
    "hidden_size": 2048,
    "norm_epsilon": 1e-06,
    "vocab_size": 151936,
    "max_position_embeddings": 8192,
    "hidden_act": "swiglu",
    "use_parallel_embedding": true,
    "embedding_sharding_dim": 0,
    "head_size": 128,
    "intermediate_size": 5632,
    "position_embedding_type": "rope_gpt_neox",
    "share_embedding_table": false,
    "residual_mlp": false,
    "bias": false,
    "rotary_pct": 1.0,
    "rank": 1,
    "decoder": "qwen",
    "rmsnorm": true,
    "lm_head_bias": false,
    "mlp_bias": false,
    "attn_bias": true,
    "rotary_base": 1000000.0,
    "rotary_scaling": null,
    "disable_weight_only_quant_plugin": false,
    "num_labels": 1,
    "use_logn_attn": false,
    "moe": {
        "num_experts": 60,
        "shared_expert_intermediate_size": 0,
        "top_k": 4,
        "normalization_mode": 0,
        "sparse_mixer_epsilon": 0.01,
        "tp_mode": 0,
        "device_limited_n_group": 0,
        "device_limited_topk_group": 0,
        "device_limited_routed_scaling_factor": 1.0
    },
    "runtime_defaults": null,
    "mapping": {
        "world_size": 2,
        "gpus_per_node": 8,
        "cp_size": 1,
        "tp_size": 2,
        "pp_size": 1,
        "moe_tp_size": 2,
        "moe_ep_size": 1,
        "auto_parallel": false
    },
    "quantization": {
        "quant_algo": "FP8",
        "kv_cache_quant_algo": null,
        "group_size": 128,
        "smoothquant_val": 0.5,
        "clamp_val": null,
        "use_meta_recipe": false,
        "has_zero_point": false,
        "pre_quant_scale": false,
        "exclude_modules": [
            "transformer.layers.7.mlp.shared_expert_gate",
            "transformer.layers.20.mlp.router",
            "lm_head",
            "transformer.layers.18.post_layernorm",
            "transformer.layers.17.mlp.shared_expert_gate",
            "transformer.layers.6.post_layernorm",
            "transformer.layers.14.mlp.router",
            "transformer.layers.23.post_layernorm",
            "transformer.layers.22.mlp.router",
            "transformer.layers.7.input_layernorm",
            "transformer.layers.9.mlp.shared_expert_gate",
            "transformer.layers.8.mlp.router",
            "transformer.layers.11.mlp.router",
            "transformer.layers.19.mlp.router",
            "transformer.layers.13.mlp.router",
            "transformer.layers.3.mlp.shared_expert_gate",
            "transformer.layers.14.post_layernorm",
            "transformer.layers.10.mlp.shared_expert_gate",
            "transformer.layers.6.input_layernorm",
            "transformer.layers.1.mlp.router",
            "transformer.layers.21.mlp.router",
            "transformer.layers.20.mlp.shared_expert_gate",
            "transformer.layers.13.input_layernorm",
            "transformer.layers.1.mlp.shared_expert_gate",
            "transformer.layers.9.input_layernorm",
            "transformer.layers.18.mlp.router",
            "transformer.layers.18.mlp.shared_expert_gate",
            "transformer.layers.21.input_layernorm",
            "transformer.layers.15.post_layernorm",
            "transformer.layers.22.mlp.shared_expert_gate",
            "transformer.layers.4.post_layernorm",
            "transformer.layers.15.input_layernorm",
            "transformer.layers.13.post_layernorm",
            "transformer.layers.4.mlp.router",
            "transformer.layers.23.mlp.router",
            "transformer.layers.3.mlp.router",
            "transformer.layers.2.mlp.router",
            "transformer.layers.21.post_layernorm",
            "transformer.layers.9.post_layernorm",
            "transformer.layers.5.mlp.shared_expert_gate",
            "transformer.layers.1.post_layernorm",
            "transformer.layers.23.input_layernorm",
            "transformer.layers.7.mlp.router",
            "transformer.layers.20.input_layernorm",
            "transformer.layers.13.mlp.shared_expert_gate",
            "transformer.layers.15.mlp.router",
            "transformer.layers.7.post_layernorm",
            "transformer.layers.22.input_layernorm",
            "transformer.layers.2.mlp.shared_expert_gate",
            "transformer.layers.17.mlp.router",
            "transformer.layers.19.post_layernorm",
            "transformer.layers.4.mlp.shared_expert_gate",
            "transformer.layers.6.mlp.shared_expert_gate",
            "transformer.layers.9.mlp.router",
            "transformer.layers.16.input_layernorm",
            "transformer.layers.0.input_layernorm",
            "transformer.layers.11.mlp.shared_expert_gate",
            "transformer.layers.12.mlp.shared_expert_gate",
            "transformer.layers.16.mlp.shared_expert_gate",
            "transformer.layers.16.post_layernorm",
            "transformer.layers.19.input_layernorm",
            "transformer.layers.14.mlp.shared_expert_gate",
            "transformer.layers.10.post_layernorm",
            "transformer.vocab_embedding",
            "transformer.layers.8.post_layernorm",
            "transformer.layers.10.input_layernorm",
            "transformer.layers.2.input_layernorm",
            "transformer.layers.14.input_layernorm",
            "transformer.layers.5.input_layernorm",
            "transformer.layers.18.input_layernorm",
            "transformer.layers.3.post_layernorm",
            "transformer.layers.5.post_layernorm",
            "transformer.layers.11.input_layernorm",
            "transformer.layers.12.input_layernorm",
            "transformer.layers.23.mlp.shared_expert_gate",
            "transformer.layers.16.mlp.router",
            "transformer.layers.15.mlp.shared_expert_gate",
            "transformer.layers.0.mlp.shared_expert_gate",
            "transformer.layers.11.post_layernorm",
            "transformer.layers.22.post_layernorm",
            "transformer.layers.10.mlp.router",
            "transformer.layers.1.input_layernorm",
            "transformer.layers.17.post_layernorm",
            "transformer.layers.12.mlp.router",
            "transformer.layers.19.mlp.shared_expert_gate",
            "transformer.layers.12.post_layernorm",
            "transformer.ln_f",
            "transformer.layers.3.input_layernorm",
            "transformer.layers.17.input_layernorm",
            "transformer.layers.8.mlp.shared_expert_gate",
            "transformer.layers.0.post_layernorm",
            "transformer.layers.0.mlp.router",
            "transformer.layers.2.post_layernorm",
            "transformer.layers.8.input_layernorm",
            "transformer.layers.4.input_layernorm",
            "transformer.layers.21.mlp.shared_expert_gate",
            "transformer.layers.6.mlp.router",
            "transformer.layers.5.mlp.router",
            "transformer.layers.20.post_layernorm"
        ]
    },
    "qk_layernorm": false,
    "rotary_embedding_dim": 128,
    "seq_length": 8192,
    "qwen_type": "qwen2_moe",
    "moe_intermediate_size": 1408,
    "moe_shared_expert_intermediate_size": 5632,
    "tie_word_embeddings": false,
    "model_type": "qwen"
}