samai-4b / config.json
tchbcb's picture
200k ctx (rope_scaling hook) + openai-compatible api (key 1234) + bootstrap v2 (chat 7861 / api 7862) + model card
f1b8f61 verified
Raw History Blame Contribute Delete
2.57 kB
{
"architectures": [
"Samai4bForCausalLM"
],
"attention_bias": false,
"attention_dropout": 0.0,
"auto_map": {
"AutoConfig": "modeling_samai_4b.Samai4bConfig",
"AutoModelForCausalLM": "modeling_samai_4b.Samai4bForCausalLM"
},
"bos_token_id": 0,
"commit_kv": true,
"drift_threshold": 0.98,
"dtype": "float16",
"eos_token_id": [
1
],
"gate_attn_act_mode": "sigmoid",
"head_dim": 256,
"headwise_attn_output_gate": true,
"hidden_act": "gelu",
"hidden_size": 2560,
"initializer_range": 0.01976,
"intermediate_size": 10240,
"layer_types": [
"sliding_attention",
"sliding_attention",
"sliding_attention",
"full_attention",
"sliding_attention",
"sliding_attention",
"sliding_attention",
"full_attention",
"sliding_attention",
"sliding_attention",
"sliding_attention",
"full_attention",
"sliding_attention",
"sliding_attention",
"sliding_attention",
"full_attention",
"sliding_attention",
"sliding_attention",
"sliding_attention",
"full_attention",
"sliding_attention",
"sliding_attention",
"sliding_attention",
"full_attention",
"sliding_attention",
"sliding_attention",
"sliding_attention",
"full_attention",
"sliding_attention",
"sliding_attention",
"sliding_attention",
"full_attention",
"sliding_attention",
"sliding_attention",
"sliding_attention",
"full_attention"
],
"max_ponder_steps": 8,
"max_position_embeddings": 200000,
"mlp_bias": false,
"model_type": "samai_4b",
"moe_aux_weight": 0.01,
"moe_num_experts": 8,
"moe_route_tau": null,
"moe_scope": "ponder",
"moe_tau": 0.125,
"mtp_loss_weight": 1.0,
"mtp_num_tokens": 2,
"num_attention_heads": 16,
"num_hidden_layers": 36,
"num_key_value_heads": 4,
"pad_token_id": 2,
"ponder_all_positions": false,
"ponder_epsilon": 0.01,
"ponder_forced_steps": 0,
"ponder_head_hidden": 0,
"ponder_loss_beta": 0.01,
"ponder_prior_p": 0.5,
"ponder_signal": "learned",
"ponder_start_layer": 28,
"ponder_step_ce_weight": 0.0,
"ponder_window": 1,
"rms_norm_eps": 1e-06,
"rope_parameters": {
"full_attention": {
"partial_rotary_factor": 0.25,
"rope_theta": 5000000.0
},
"rope_theta": 10000.0,
"rope_type": "default",
"sliding_attention": {
"partial_rotary_factor": 1.0,
"rope_theta": 10000.0
}
},
"sliding_window": 512,
"tie_word_embeddings": true,
"transformers_version": "5.16.1",
"use_cache": true,
"vocab_size": 131072
}