Hugging Face's logo

runanywhere
/
phi_tiny_moe_HNPU

Text Generation
hnpu
hexagon
npu
llm
v79
v81
Model card Files
xet
 Community
Aman0runanywhere commited on
Commit
2fbc78b
·
verified ·
1 Parent(s): 3b770d6

Add chat template block (runtime-owned chat templating)

Browse files
Files changed (1) hide show
  1. v79/phimoe.json +85 -31
v79/phimoe.json CHANGED
@@ -1,31 +1,85 @@
1
- {
2
- "schema_version": 1,
3
- "_comment": "phi-tiny-moe - Qualcomm Hexagon v79 (QHexRT) deployment manifest.",
4
- "model": { "name": "phi-tiny-moe", "family": "llm", "dsp_arch": "v79", "tokenizer_pre": "default" },
5
- "params": {
6
- "hidden": 4096, "vocab": 32064, "n_layers": 32, "max_ctx": 2048,
7
- "kv_dim": 512, "head_dim": 128, "rope_theta": 10000, "eos_token_id": 32000
8
- },
9
- "artifacts": {
10
- "contexts": {
11
- "a_lo": { "bin": "a_lo.bin" },
12
- "a_hi": { "bin": "a_hi.bin" },
13
- "ffn": { "bin": "ffn2.bin" },
14
- "lmhead": { "bin": "lmhead_ar1.bin" },
15
- "pf_lo": { "bin": "pf_lo.bin" },
16
- "pf_hi": { "bin": "pf_hi.bin" },
17
- "ffn_pf": { "bin": "ffn_pf.bin" }
18
- },
19
- "embed": "embed_f16.bin",
20
- "tokenizer": "tokenizer.json",
21
- "fixture_dir": "."
22
- },
23
- "plan": {
24
- "steps": [
25
- { "host": "phimoe_generate",
26
- "params": { "experts": "experts_i8.bin", "scales": "experts_scale.f32",
27
- "n_experts": 16, "ffn": 448, "jitter": 0.01, "neg": -50,
28
- "user_id": 32010, "end_id": 32007, "assistant_id": 32001, "max_new": 24 } }
29
- ]
30
- }
31
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "schema_version": 1,
3
+ "_comment": "phi-tiny-moe - Qualcomm Hexagon v79 (QHexRT) deployment manifest.",
4
+ "model": {
5
+ "name": "phi-tiny-moe",
6
+ "family": "llm",
7
+ "dsp_arch": "v79",
8
+ "tokenizer_pre": "default"
9
+ },
10
+ "chat": {
11
+ "user": {
12
+ "prefix": [
13
+ 32010
14
+ ],
15
+ "suffix": [
16
+ 32007
17
+ ]
18
+ },
19
+ "assistant": {
20
+ "prefix": [
21
+ 32001
22
+ ],
23
+ "suffix": [
24
+ 32007
25
+ ]
26
+ }
27
+ },
28
+ "params": {
29
+ "hidden": 4096,
30
+ "vocab": 32064,
31
+ "n_layers": 32,
32
+ "max_ctx": 2048,
33
+ "kv_dim": 512,
34
+ "head_dim": 128,
35
+ "rope_theta": 10000,
36
+ "eos_token_id": 32000
37
+ },
38
+ "artifacts": {
39
+ "contexts": {
40
+ "a_lo": {
41
+ "bin": "a_lo.bin"
42
+ },
43
+ "a_hi": {
44
+ "bin": "a_hi.bin"
45
+ },
46
+ "ffn": {
47
+ "bin": "ffn2.bin"
48
+ },
49
+ "lmhead": {
50
+ "bin": "lmhead_ar1.bin"
51
+ },
52
+ "pf_lo": {
53
+ "bin": "pf_lo.bin"
54
+ },
55
+ "pf_hi": {
56
+ "bin": "pf_hi.bin"
57
+ },
58
+ "ffn_pf": {
59
+ "bin": "ffn_pf.bin"
60
+ }
61
+ },
62
+ "embed": "embed_f16.bin",
63
+ "tokenizer": "tokenizer.json",
64
+ "fixture_dir": "."
65
+ },
66
+ "plan": {
67
+ "steps": [
68
+ {
69
+ "host": "phimoe_generate",
70
+ "params": {
71
+ "experts": "experts_i8.bin",
72
+ "scales": "experts_scale.f32",
73
+ "n_experts": 16,
74
+ "ffn": 448,
75
+ "jitter": 0.01,
76
+ "neg": -50,
77
+ "user_id": 32010,
78
+ "end_id": 32007,
79
+ "assistant_id": 32001,
80
+ "max_new": 24
81
+ }
82
+ }
83
+ ]
84
+ }
85
+ }