ansblnet/swarm / config.json
ansblnet's picture
download
raw
3.13 kB
{
"architectures": [
"LimiteForCausalLM"
],
"auto_map": {
"AutoConfig": "configuration_limite.LimiteConfig",
"AutoModel": "modeling_limite.LimiteModel",
"AutoModelForCausalLM": "modeling_limite.LimiteForCausalLM"
},
"attention_softmax_scale": 0.1,
"attn_gate_applied": "per_head_before_o_proj",
"attn_gate_channels": 128,
"attn_gate_scale": 2.0,
"bos_token_id": 151643,
"checkpoint_step": -1,
"checkpoint_world_size": 48,
"eos_token_id": 151643,
"final_softcap": 0.0,
"global_every": 4,
"global_layers": [
3,
7,
11,
15,
19,
23,
27,
31,
35,
39,
43,
47
],
"global_nope": true,
"global_window": -1,
"head_dim": 128,
"hidden_size": 1280,
"intermediate_size": 3328,
"lm_head_precision_mode": "oracle_exact",
"max_position_embeddings": 131072,
"mlp_formula": "(silu(gate_proj(h)) * up_proj(h)) @ down_proj, no clamp on either factor",
"mlp_ratio": 4,
"mlp_type": "swiglu",
"model_type": "limite",
"mudd": true,
"mudd_accumulation": "ordered_left_to_right",
"mudd_at": [
24,
47
],
"mudd_hist_convention": "hist[0] = rms_norm(embedding); hist[j] = output of layer j-1",
"mudd_inter": 32,
"mudd_layers": [
24,
47
],
"mudd_mlp": true,
"mudd_r_site": "resid",
"mudd_tap_idx": {
"24": [
0,
12,
24
],
"47": [
0,
23,
47
]
},
"mudd_taps": 3,
"num_attention_heads": 10,
"num_hidden_layers": 48,
"num_key_value_heads": 2,
"pad_token_id": 151643,
"padded_vocab_size": 151680,
"pos_mode": "rope",
"qk_norm": "rms_pre_rope",
"rms_norm_eps_mode": "torch_finfo_default",
"rms_norm_has_weight": false,
"rope_base_global": 1024.0,
"rope_base_local": 1024.0,
"rope_cos_sin_dtype": "bfloat16",
"rope_frac": 0.5,
"rope_n_pairs": 32,
"rope_per_layer": false,
"rope_style": "interleaved_pairs_odd_lane_sign_flip",
"sliding_window": 1024,
"sliding_window_convention": "k >= q - sliding_window, inclusive of the query token (span = sliding_window + 1 keys)",
"softcap_logits": {
"a": 23.0,
"b": 5.0,
"c": 7.5,
"kind": "sigmoid"
},
"source_format": "guido-export-v1",
"source_metadata_assumptions": [],
"tie_word_embeddings": true,
"tokenizer_vocab_size": 151667,
"torch_dtype": "bfloat16",
"ve_applied_before_qk_norm": true,
"ve_dim": 128,
"ve_gate_channels": 12,
"ve_gate_scale": 2.0,
"ve_head_slice": "first_num_key_value_heads",
"ve_layers": [
1,
4,
7,
10,
13,
16,
19,
22,
25,
28,
31,
34,
37,
40,
43,
46
],
"ve_stored_heads": 2,
"vocab_size": 151680,
"xsa": true,
"xsa_layers": [
0,
1,
2,
3,
4,
5,
6,
7,
8,
9,
10,
11,
12,
13,
14,
15,
16,
17,
18,
19,
20,
21,
22,
23,
24,
25,
26,
27,
28,
29,
30,
31,
32,
33,
34,
35,
36,
37,
38,
39,
40,
41,
42,
43,
44,
45,
46,
47
],
"xsa_normalize_eps": 0.0001
}

Xet Storage Details

Size:
3.13 kB
·
Xet hash:
ae069528270ab3eaa86612030e616bf7a3eb9952617761a460cf15977e80c8f3

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.