sarvam-30b / config.json
matrixdose's picture
Duplicate from sarvamai/sarvam-30b
c31cf57
{
"architectures": [
"SarvamMoEForCausalLM"
],
"attention_dropout": 0.0,
"attn_implementation": null,
"auto_map": {
"AutoConfig": "configuration_sarvam_moe.SarvamMoEConfig",
"AutoModel": "modeling_sarvam_moe.SarvamMoEModel",
"AutoModelForCausalLM": "modeling_sarvam_moe.SarvamMoEForCausalLM"
},
"dtype": "float32",
"embedding_dropout": 0.0,
"eos_token_id": 1,
"first_k_dense_replace": 1,
"head_dim": 64,
"hidden_act": "silu",
"hidden_size": 4096,
"initializer_range": 0.006,
"intermediate_size": 8192,
"max_position_embeddings": 131072,
"max_window_layers": 19,
"model_type": "sarvam_moe",
"moe_intermediate_size": 1024,
"moe_router_enable_expert_bias": true,
"moe_shared_expert_intermediate_size": 1024,
"n_group": 1,
"norm_topk_prob": true,
"num_attention_heads": 64,
"num_experts": 128,
"num_experts_per_tok": 6,
"num_hidden_layers": 19,
"num_key_value_heads": 4,
"num_shared_experts": 1,
"output_dropout": 0.0,
"output_router_logits": false,
"pad_token_id": 0,
"rms_norm_eps": 1e-06,
"rope_scaling": null,
"rope_theta": 8000000,
"routed_scaling_factor": 2.5,
"router_dtype": "fp32",
"score_function": "sigmoid",
"tie_word_embeddings": false,
"topk_group": 1,
"transformers_version": "4.57.2",
"use_bias": false,
"use_cache": true,
"use_qk_norm": true,
"use_qkv_bias": false,
"use_rmsnorm": true,
"vocab_size": 262144
}