Files
convergent-mamba2-300M-adam…/config.json
ModelHub XC b03c9fad5c 初始化项目,由ModelHub XC社区提供模型
Model: deqing/convergent-mamba2-300M-adamw-original
Source: Original Platform
2026-08-21 02:43:16 +08:00

46 lines
1003 B
JSON

{
"architectures": [
"Mamba2ForCausalLM"
],
"bos_token_id": 128000,
"chunk_size": 256,
"conv_kernel": 4,
"dtype": "float32",
"eos_token_id": 128001,
"expand": 2,
"fuse_cross_entropy": true,
"fuse_linear_cross_entropy": false,
"fuse_norm": true,
"head_dim": 64,
"hidden_act": "silu",
"hidden_size": 1024,
"initializer_range": 0.02,
"layer_norm_epsilon": 1e-05,
"max_position_embeddings": 1024,
"model_type": "mamba2",
"n_groups": 1,
"norm_eps": 1e-05,
"num_heads": 32,
"num_hidden_layers": 28,
"pad_token_id": null,
"rescale_prenorm_residual": true,
"residual_in_fp32": true,
"rms_norm": true,
"state_size": 128,
"tie_word_embeddings": true,
"time_step_floor": 0.0001,
"time_step_limit": [
0.0,
1e+30
],
"time_step_max": 0.1,
"time_step_min": 0.001,
"time_step_rank": 128,
"transformers_version": "5.3.0",
"use_bias": false,
"use_cache": false,
"use_conv_bias": true,
"use_l2warp": false,
"vocab_size": 128256
}