Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,134 @@
|
|
1 |
-
---
|
2 |
-
|
3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
pipeline_tag: text-generation
|
3 |
+
inference: false
|
4 |
+
license: apache-2.0
|
5 |
+
model-index:
|
6 |
+
- name: ibm/PowerMoE-3b
|
7 |
+
results:
|
8 |
+
- task:
|
9 |
+
type: text-generation
|
10 |
+
dataset:
|
11 |
+
type: lm-eval-harness
|
12 |
+
name: ARC
|
13 |
+
metrics:
|
14 |
+
- name: accuracy-norm
|
15 |
+
type: accuracy-norm
|
16 |
+
value: 58.1
|
17 |
+
verified: false
|
18 |
+
- task:
|
19 |
+
type: text-generation
|
20 |
+
dataset:
|
21 |
+
type: lm-eval-harness
|
22 |
+
name: BoolQ
|
23 |
+
metrics:
|
24 |
+
- name: accuracy
|
25 |
+
type: accuracy
|
26 |
+
value: 65
|
27 |
+
verified: false
|
28 |
+
- task:
|
29 |
+
type: text-generation
|
30 |
+
dataset:
|
31 |
+
type: lm-eval-harness
|
32 |
+
name: Hellaswag
|
33 |
+
metrics:
|
34 |
+
- name: accuracy-norm
|
35 |
+
type: accuracy-norm
|
36 |
+
value: 71.5
|
37 |
+
verified: false
|
38 |
+
- task:
|
39 |
+
type: text-generation
|
40 |
+
dataset:
|
41 |
+
type: lm-eval-harness
|
42 |
+
name: OpenBookQA
|
43 |
+
metrics:
|
44 |
+
- name: accuracy-norm
|
45 |
+
type: accuracy-norm
|
46 |
+
value: 41
|
47 |
+
verified: false
|
48 |
+
- task:
|
49 |
+
type: text-generation
|
50 |
+
dataset:
|
51 |
+
type: lm-eval-harness
|
52 |
+
name: PIQA
|
53 |
+
metrics:
|
54 |
+
- name: accuracy-norm
|
55 |
+
type: accuracy-norm
|
56 |
+
value: 79.1
|
57 |
+
verified: false
|
58 |
+
- task:
|
59 |
+
type: text-generation
|
60 |
+
dataset:
|
61 |
+
type: lm-eval-harness
|
62 |
+
name: Winogrande
|
63 |
+
metrics:
|
64 |
+
- name: accuracy-norm
|
65 |
+
type: accuracy-norm
|
66 |
+
value: 65
|
67 |
+
verified: false
|
68 |
+
- task:
|
69 |
+
type: text-generation
|
70 |
+
dataset:
|
71 |
+
type: lm-eval-harness
|
72 |
+
name: MMLU (5 shot)
|
73 |
+
metrics:
|
74 |
+
- name: accuracy
|
75 |
+
type: accuracy
|
76 |
+
value: 42.8
|
77 |
+
verified: false
|
78 |
+
- task:
|
79 |
+
type: text-generation
|
80 |
+
dataset:
|
81 |
+
type: lm-eval-harness
|
82 |
+
name: GSM8k (5 shot)
|
83 |
+
metrics:
|
84 |
+
- name: accuracy
|
85 |
+
type: accuracy
|
86 |
+
value: 25.9
|
87 |
+
verified: false
|
88 |
+
- task:
|
89 |
+
type: text-generation
|
90 |
+
dataset:
|
91 |
+
type: lm-eval-harness
|
92 |
+
name: math (4 shot)
|
93 |
+
metrics:
|
94 |
+
- name: accuracy
|
95 |
+
type: accuracy
|
96 |
+
value: 14.8
|
97 |
+
verified: false
|
98 |
+
- task:
|
99 |
+
type: text-generation
|
100 |
+
dataset:
|
101 |
+
type: bigcode-eval
|
102 |
+
name: humaneval
|
103 |
+
metrics:
|
104 |
+
- name: pass@1
|
105 |
+
type: pass@1
|
106 |
+
value: 20.1
|
107 |
+
verified: false
|
108 |
+
- task:
|
109 |
+
type: text-generation
|
110 |
+
dataset:
|
111 |
+
type: bigcode-eval
|
112 |
+
name: MBPP
|
113 |
+
metrics:
|
114 |
+
- name: pass@1
|
115 |
+
type: pass@1
|
116 |
+
value: 32.4
|
117 |
+
verified: false
|
118 |
+
base_model:
|
119 |
+
- ibm/PowerMoE-3b
|
120 |
+
---
|
121 |
+
|
122 |
+
## Model Summary
|
123 |
+
PowerMoE-3B is a 3B sparse Mixture-of-Experts (sMoE) language model trained with the Power learning rate scheduler. It sparsely activates 800M parameters for each token. It is trained on a mix of open-source and proprietary datasets. PowerMoE-3B has shown promising results compared to other dense models with 2x activate parameters across various benchmarks, including natural language multi-choices, code generation, and math reasoning.
|
124 |
+
Paper: https://arxiv.org/abs/2408.13359
|
125 |
+
|
126 |
+
This is a GGUF quantized version.
|
127 |
+
|
128 |
+
## Usage
|
129 |
+
Requires latest llama.cpp to run.
|
130 |
+
|
131 |
+
### Generation
|
132 |
+
This is a simple example of how to use the PowerMoe GGUF:
|
133 |
+
|
134 |
+
./llama-cli -m PowerMoE4x800M_q3km.gguf -p "How about a snack?"
|