flint:40b-a2b-mlx-bf16

2 13 hours ago

Flint is a 39.8 billion-parameter mixture-of-experts model from Recursive, built for coding agents.

tools thinking 40b
... /
json
5fa9d05d3ef3 ยท 2.5kB
{
"architectures": [
"FlintForCausalLM"
],
"model_type": "flint",
"auto_map": {
"AutoConfig": "configuration_flint.FlintConfig"
},
"attention_bias": false,
"attention_dropout": 0.0,
"bos_token_id": 150000,
"decoder_sparse_step": 1,
"eos_token_id": [
150001,
150003,
150005
],
"head_dim": 128,
"hidden_act": "silu",
"hidden_size": 2048,
"initializer_range": 0.02,
"intermediate_size": 6144,
"layer_types": [
"full_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",
"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",
"full_attention"
],
"max_position_embeddings": 262144,
"mlp_only_layers": [
0,
1
],
"moe_intermediate_size": 512,
"norm_topk_prob": true,
"num_attention_heads": 32,
"num_experts": 256,
"num_experts_per_tok": 6,
"num_hidden_layers": 49,
"num_key_value_heads": 4,
"no_rope_layer_indices": [
0,
1,
5,
9,
13,
17,
21,
25,
29,
33,
37,
41,
45,
48
],
"output_router_logits": false,
"pad_token_id": 150001,
"rms_norm_eps": 1e-06,
"rope_theta": 10000.0,
"router_aux_loss_coef": 0.0005,
"router_score_function": "sigmoid",
"shared_expert_intermediate_size": 1024,
"sliding_window": 513,
"softmax_type": "learnable",
"tie_word_embeddings": false,
"torch_dtype": "bfloat16",
"use_cache": true,
"use_sliding_window": true,
"vocab_size": 150144
}