AbstractPhil commited on
Commit
56d242f
·
verified ·
1 Parent(s): 4296fa7

bed: alhena-both_s1_L24000_fl0.8_ctx4096_frdyn_cap_mode_sz32-128-256-512_r4_n30_o0_askany_p75d_gag50_cp4_gc96_ac44_kh50_va1_pt10b4h0_krt16r218_ahdigit1_code44sa weights

Browse files
bed/weights/alhena-both_s1_L24000_fl0.8_ctx4096_frdyn_cap_mode_sz32-128-256-512_r4_n30_o0_askany_p75d_gag50_cp4_gc96_ac44_kh50_va1_pt10b4h0_krt16r218_ahdigit1_code44sa/addr_head.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f16600945f5f6d169cf7d10916a6d2cc95dbfb8d58779d8c9181b0c131232125
3
+ size 16584
bed/weights/alhena-both_s1_L24000_fl0.8_ctx4096_frdyn_cap_mode_sz32-128-256-512_r4_n30_o0_askany_p75d_gag50_cp4_gc96_ac44_kh50_va1_pt10b4h0_krt16r218_ahdigit1_code44sa/config.json ADDED
@@ -0,0 +1,46 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "bed-alhena-both",
3
+ "d_model": 256,
4
+ "n_layers": 6,
5
+ "n_heads": 8,
6
+ "context": 4096,
7
+ "vocab_size": 256,
8
+ "tokenizer": "byte-trigram",
9
+ "hub_layers": [],
10
+ "hub_K": 32,
11
+ "hub_D": 16,
12
+ "tau": 0.1,
13
+ "bank_experts": 7,
14
+ "bank_ff": 256,
15
+ "head_K": 64,
16
+ "head_D": 16,
17
+ "head_mode": "",
18
+ "head_dial_band": 4.0,
19
+ "head_dial_rho": 0.015625,
20
+ "gate_init": -3.0,
21
+ "tie_embeddings": false,
22
+ "hub_chunk": 128,
23
+ "hub_const": 1,
24
+ "hub_ckpt": 0,
25
+ "fusion": null,
26
+ "qk_norm": "",
27
+ "attn_fp32": false,
28
+ "attn_kernel": "sdpa",
29
+ "lookup_arm_sites": [],
30
+ "lookup_arm_depth": 1,
31
+ "attn_rotary": true,
32
+ "hub_rotary": false,
33
+ "rotary_base": 10000.0,
34
+ "hub_rotary_frac": 1.0,
35
+ "hub_dual": false,
36
+ "pos_table": true,
37
+ "bank_byte_table": "add",
38
+ "bank_last_fix": "",
39
+ "fix_rho": 0.0625,
40
+ "fix_warm": 200,
41
+ "fix_clamp": [
42
+ 0.25,
43
+ 4.0
44
+ ],
45
+ "canon": ""
46
+ }
bed/weights/alhena-both_s1_L24000_fl0.8_ctx4096_frdyn_cap_mode_sz32-128-256-512_r4_n30_o0_askany_p75d_gag50_cp4_gc96_ac44_kh50_va1_pt10b4h0_krt16r218_ahdigit1_code44sa/key_rot.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b0d0beb49e4ad077b36d3eb5a3fdb41767146904500db46503b528ae751491e1
3
+ size 147536
bed/weights/alhena-both_s1_L24000_fl0.8_ctx4096_frdyn_cap_mode_sz32-128-256-512_r4_n30_o0_askany_p75d_gag50_cp4_gc96_ac44_kh50_va1_pt10b4h0_krt16r218_ahdigit1_code44sa/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cc8ff365e6fda806986cedf171799bbd652f7a04ab55b242a7e14e148c1c0f22
3
+ size 36916028
bed/weights/alhena-both_s1_L24000_fl0.8_ctx4096_frdyn_cap_mode_sz32-128-256-512_r4_n30_o0_askany_p75d_gag50_cp4_gc96_ac44_kh50_va1_pt10b4h0_krt16r218_ahdigit1_code44sa/scale.json ADDED
@@ -0,0 +1,84 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "grid": "ag",
3
+ "grid_p": 0.5,
4
+ "n_ag": 4,
5
+ "rotary": {
6
+ "form": "anchored",
7
+ "pairs": 16,
8
+ "rel_pairs": 4,
9
+ "content_pairs": 4,
10
+ "anchor_freqs": 2,
11
+ "levels": 4,
12
+ "ntk_scale": 4.0,
13
+ "base": 10000.0,
14
+ "null": 4,
15
+ "w": [
16
+ 1.0,
17
+ 0.06299605249474366,
18
+ 0.003968502629920499,
19
+ 0.00025,
20
+ 0.0,
21
+ 0.0,
22
+ 0.0,
23
+ 0.0,
24
+ 1.2566370614359172,
25
+ 2.5132741228718345,
26
+ 1.2566370614359172,
27
+ 2.5132741228718345,
28
+ 1.2566370614359172,
29
+ 2.5132741228718345,
30
+ 1.2566370614359172,
31
+ 2.5132741228718345
32
+ ],
33
+ "kind": [
34
+ 0,
35
+ 0,
36
+ 0,
37
+ 0,
38
+ 1,
39
+ 1,
40
+ 1,
41
+ 1,
42
+ 2,
43
+ 2,
44
+ 3,
45
+ 3,
46
+ 4,
47
+ 4,
48
+ 5,
49
+ 5
50
+ ],
51
+ "code": "4x4"
52
+ },
53
+ "head": "digit",
54
+ "head_lam": 1.0,
55
+ "layout": {
56
+ "slot": 32,
57
+ "comps": 96,
58
+ "radix": 4,
59
+ "levels": 4
60
+ },
61
+ "grid_code": "4x4",
62
+ "grid_layout": "AG96",
63
+ "grid_hard": 0.5,
64
+ "value_addr": {
65
+ "levels": 4,
66
+ "digits": 4,
67
+ "d": 256
68
+ },
69
+ "key_rot": {
70
+ "pairs": 16,
71
+ "k": 9,
72
+ "vocab": 256,
73
+ "block": 4,
74
+ "head": 0,
75
+ "seed_tag": "KEYROT|v1|j16",
76
+ "norm": 2,
77
+ "rho": 2.18
78
+ },
79
+ "ptr": {
80
+ "lam": 0.1,
81
+ "block": 4,
82
+ "head": 0
83
+ }
84
+ }
bed/weights/alhena-both_s1_L24000_fl0.8_ctx4096_frdyn_cap_mode_sz32-128-256-512_r4_n30_o0_askany_p75d_gag50_cp4_gc96_ac44_kh50_va1_pt10b4h0_krt16r218_ahdigit1_code44sa/value_addr.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3fcb60711424cc8b3900114eea351371e025b4aaf2bb7d2b94096cfcf1bf6d26
3
+ size 16464