Commit
•
7594795
1
Parent(s):
a43e490
uploaded readme
Browse files
README.md
ADDED
@@ -0,0 +1,186 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
Quantization made by Richard Erkhov.
|
2 |
+
|
3 |
+
[Github](https://github.com/RichardErkhov)
|
4 |
+
|
5 |
+
[Discord](https://discord.gg/pvy7H8DZMG)
|
6 |
+
|
7 |
+
[Request more models](https://github.com/RichardErkhov/quant_request)
|
8 |
+
|
9 |
+
|
10 |
+
gemma-2b-zephyr-sft - GGUF
|
11 |
+
- Model creator: https://huggingface.co/Columbia-NLP/
|
12 |
+
- Original model: https://huggingface.co/Columbia-NLP/gemma-2b-zephyr-sft/
|
13 |
+
|
14 |
+
|
15 |
+
| Name | Quant method | Size |
|
16 |
+
| ---- | ---- | ---- |
|
17 |
+
| [gemma-2b-zephyr-sft.Q2_K.gguf](https://huggingface.co/RichardErkhov/Columbia-NLP_-_gemma-2b-zephyr-sft-gguf/blob/main/gemma-2b-zephyr-sft.Q2_K.gguf) | Q2_K | 1.08GB |
|
18 |
+
| [gemma-2b-zephyr-sft.IQ3_XS.gguf](https://huggingface.co/RichardErkhov/Columbia-NLP_-_gemma-2b-zephyr-sft-gguf/blob/main/gemma-2b-zephyr-sft.IQ3_XS.gguf) | IQ3_XS | 1.16GB |
|
19 |
+
| [gemma-2b-zephyr-sft.IQ3_S.gguf](https://huggingface.co/RichardErkhov/Columbia-NLP_-_gemma-2b-zephyr-sft-gguf/blob/main/gemma-2b-zephyr-sft.IQ3_S.gguf) | IQ3_S | 1.2GB |
|
20 |
+
| [gemma-2b-zephyr-sft.Q3_K_S.gguf](https://huggingface.co/RichardErkhov/Columbia-NLP_-_gemma-2b-zephyr-sft-gguf/blob/main/gemma-2b-zephyr-sft.Q3_K_S.gguf) | Q3_K_S | 1.2GB |
|
21 |
+
| [gemma-2b-zephyr-sft.IQ3_M.gguf](https://huggingface.co/RichardErkhov/Columbia-NLP_-_gemma-2b-zephyr-sft-gguf/blob/main/gemma-2b-zephyr-sft.IQ3_M.gguf) | IQ3_M | 1.22GB |
|
22 |
+
| [gemma-2b-zephyr-sft.Q3_K.gguf](https://huggingface.co/RichardErkhov/Columbia-NLP_-_gemma-2b-zephyr-sft-gguf/blob/main/gemma-2b-zephyr-sft.Q3_K.gguf) | Q3_K | 1.29GB |
|
23 |
+
| [gemma-2b-zephyr-sft.Q3_K_M.gguf](https://huggingface.co/RichardErkhov/Columbia-NLP_-_gemma-2b-zephyr-sft-gguf/blob/main/gemma-2b-zephyr-sft.Q3_K_M.gguf) | Q3_K_M | 1.29GB |
|
24 |
+
| [gemma-2b-zephyr-sft.Q3_K_L.gguf](https://huggingface.co/RichardErkhov/Columbia-NLP_-_gemma-2b-zephyr-sft-gguf/blob/main/gemma-2b-zephyr-sft.Q3_K_L.gguf) | Q3_K_L | 1.36GB |
|
25 |
+
| [gemma-2b-zephyr-sft.IQ4_XS.gguf](https://huggingface.co/RichardErkhov/Columbia-NLP_-_gemma-2b-zephyr-sft-gguf/blob/main/gemma-2b-zephyr-sft.IQ4_XS.gguf) | IQ4_XS | 1.4GB |
|
26 |
+
| [gemma-2b-zephyr-sft.Q4_0.gguf](https://huggingface.co/RichardErkhov/Columbia-NLP_-_gemma-2b-zephyr-sft-gguf/blob/main/gemma-2b-zephyr-sft.Q4_0.gguf) | Q4_0 | 1.44GB |
|
27 |
+
| [gemma-2b-zephyr-sft.IQ4_NL.gguf](https://huggingface.co/RichardErkhov/Columbia-NLP_-_gemma-2b-zephyr-sft-gguf/blob/main/gemma-2b-zephyr-sft.IQ4_NL.gguf) | IQ4_NL | 1.45GB |
|
28 |
+
| [gemma-2b-zephyr-sft.Q4_K_S.gguf](https://huggingface.co/RichardErkhov/Columbia-NLP_-_gemma-2b-zephyr-sft-gguf/blob/main/gemma-2b-zephyr-sft.Q4_K_S.gguf) | Q4_K_S | 1.45GB |
|
29 |
+
| [gemma-2b-zephyr-sft.Q4_K.gguf](https://huggingface.co/RichardErkhov/Columbia-NLP_-_gemma-2b-zephyr-sft-gguf/blob/main/gemma-2b-zephyr-sft.Q4_K.gguf) | Q4_K | 1.52GB |
|
30 |
+
| [gemma-2b-zephyr-sft.Q4_K_M.gguf](https://huggingface.co/RichardErkhov/Columbia-NLP_-_gemma-2b-zephyr-sft-gguf/blob/main/gemma-2b-zephyr-sft.Q4_K_M.gguf) | Q4_K_M | 1.52GB |
|
31 |
+
| [gemma-2b-zephyr-sft.Q4_1.gguf](https://huggingface.co/RichardErkhov/Columbia-NLP_-_gemma-2b-zephyr-sft-gguf/blob/main/gemma-2b-zephyr-sft.Q4_1.gguf) | Q4_1 | 1.56GB |
|
32 |
+
| [gemma-2b-zephyr-sft.Q5_0.gguf](https://huggingface.co/RichardErkhov/Columbia-NLP_-_gemma-2b-zephyr-sft-gguf/blob/main/gemma-2b-zephyr-sft.Q5_0.gguf) | Q5_0 | 1.68GB |
|
33 |
+
| [gemma-2b-zephyr-sft.Q5_K_S.gguf](https://huggingface.co/RichardErkhov/Columbia-NLP_-_gemma-2b-zephyr-sft-gguf/blob/main/gemma-2b-zephyr-sft.Q5_K_S.gguf) | Q5_K_S | 1.68GB |
|
34 |
+
| [gemma-2b-zephyr-sft.Q5_K.gguf](https://huggingface.co/RichardErkhov/Columbia-NLP_-_gemma-2b-zephyr-sft-gguf/blob/main/gemma-2b-zephyr-sft.Q5_K.gguf) | Q5_K | 1.71GB |
|
35 |
+
| [gemma-2b-zephyr-sft.Q5_K_M.gguf](https://huggingface.co/RichardErkhov/Columbia-NLP_-_gemma-2b-zephyr-sft-gguf/blob/main/gemma-2b-zephyr-sft.Q5_K_M.gguf) | Q5_K_M | 1.71GB |
|
36 |
+
| [gemma-2b-zephyr-sft.Q5_1.gguf](https://huggingface.co/RichardErkhov/Columbia-NLP_-_gemma-2b-zephyr-sft-gguf/blob/main/gemma-2b-zephyr-sft.Q5_1.gguf) | Q5_1 | 1.79GB |
|
37 |
+
| [gemma-2b-zephyr-sft.Q6_K.gguf](https://huggingface.co/RichardErkhov/Columbia-NLP_-_gemma-2b-zephyr-sft-gguf/blob/main/gemma-2b-zephyr-sft.Q6_K.gguf) | Q6_K | 1.92GB |
|
38 |
+
| [gemma-2b-zephyr-sft.Q8_0.gguf](https://huggingface.co/RichardErkhov/Columbia-NLP_-_gemma-2b-zephyr-sft-gguf/blob/main/gemma-2b-zephyr-sft.Q8_0.gguf) | Q8_0 | 2.49GB |
|
39 |
+
|
40 |
+
|
41 |
+
|
42 |
+
|
43 |
+
Original model description:
|
44 |
+
---
|
45 |
+
license: other
|
46 |
+
license_name: gemma-terms-of-use
|
47 |
+
license_link: https://ai.google.dev/gemma/terms
|
48 |
+
base_model: google/gemma-2b
|
49 |
+
tags:
|
50 |
+
- alignment-handbook
|
51 |
+
- trl
|
52 |
+
- sft
|
53 |
+
- generated_from_trainer
|
54 |
+
datasets:
|
55 |
+
- HuggingFaceH4/deita-10k-v0-sft
|
56 |
+
model-index:
|
57 |
+
- name: gemma-2b-zephyr-sft
|
58 |
+
results:
|
59 |
+
- task:
|
60 |
+
type: text-generation
|
61 |
+
name: Text Generation
|
62 |
+
dataset:
|
63 |
+
name: AI2 Reasoning Challenge (25-Shot)
|
64 |
+
type: ai2_arc
|
65 |
+
config: ARC-Challenge
|
66 |
+
split: test
|
67 |
+
args:
|
68 |
+
num_few_shot: 25
|
69 |
+
metrics:
|
70 |
+
- type: acc_norm
|
71 |
+
value: 51.88
|
72 |
+
name: normalized accuracy
|
73 |
+
- task:
|
74 |
+
type: text-generation
|
75 |
+
name: Text Generation
|
76 |
+
dataset:
|
77 |
+
name: HellaSwag (10-Shot)
|
78 |
+
type: hellaswag
|
79 |
+
split: validation
|
80 |
+
args:
|
81 |
+
num_few_shot: 10
|
82 |
+
metrics:
|
83 |
+
- type: acc_norm
|
84 |
+
value: 72.63
|
85 |
+
name: normalized accuracy
|
86 |
+
- task:
|
87 |
+
type: text-generation
|
88 |
+
name: Text Generation
|
89 |
+
dataset:
|
90 |
+
name: MMLU (5-Shot)
|
91 |
+
type: cais/mmlu
|
92 |
+
config: all
|
93 |
+
split: test
|
94 |
+
args:
|
95 |
+
num_few_shot: 5
|
96 |
+
metrics:
|
97 |
+
- type: acc
|
98 |
+
value: 42.20
|
99 |
+
name: accuracy
|
100 |
+
- task:
|
101 |
+
type: text-generation
|
102 |
+
name: Text Generation
|
103 |
+
dataset:
|
104 |
+
name: TruthfulQA (0-shot)
|
105 |
+
type: truthful_qa
|
106 |
+
config: multiple_choice
|
107 |
+
split: validation
|
108 |
+
args:
|
109 |
+
num_few_shot: 0
|
110 |
+
metrics:
|
111 |
+
- type: mc2
|
112 |
+
value: 41.96
|
113 |
+
- task:
|
114 |
+
type: text-generation
|
115 |
+
name: Text Generation
|
116 |
+
dataset:
|
117 |
+
name: Winogrande (5-shot)
|
118 |
+
type: winogrande
|
119 |
+
config: winogrande_xl
|
120 |
+
split: validation
|
121 |
+
args:
|
122 |
+
num_few_shot: 5
|
123 |
+
metrics:
|
124 |
+
- type: acc
|
125 |
+
value: 63.85
|
126 |
+
name: accuracy
|
127 |
+
- task:
|
128 |
+
type: text-generation
|
129 |
+
name: Text Generation
|
130 |
+
dataset:
|
131 |
+
name: GSM8k (5-shot)
|
132 |
+
type: gsm8k
|
133 |
+
config: main
|
134 |
+
split: test
|
135 |
+
args:
|
136 |
+
num_few_shot: 5
|
137 |
+
metrics:
|
138 |
+
- type: acc
|
139 |
+
value: 20.09
|
140 |
+
name: accuracy
|
141 |
+
---
|
142 |
+
|
143 |
+
# Model Card for Gemma 2B Zephyr SFT
|
144 |
+
|
145 |
+
We trained the [google/gemma-2b](https://huggingface.co/google/gemma-2b) with [deita-10k-v0-sft](https://huggingface.co/datasets/HuggingFaceH4/deita-10k-v0-sft).
|
146 |
+
We carefully selected the hyper-parameters and masked the user tokens during training to achieve the best supervised fine-tuning performance.
|
147 |
+
|
148 |
+
## Model description
|
149 |
+
|
150 |
+
- **Model type:** A 2.5B parameter GPT-like model fine-tuned on a mix of publicly available, synthetic datasets.
|
151 |
+
- **Language(s) (NLP):** Primarily English
|
152 |
+
- **License:** Gemma Terms of Use
|
153 |
+
- **Finetuned from model:** [google/gemma-2b](https://huggingface.co/google/gemma-2b)
|
154 |
+
|
155 |
+
|
156 |
+
## License
|
157 |
+
This model has the same license as the [original Gemma model collection](https://ai.google.dev/gemma/terms)
|
158 |
+
|
159 |
+
## OpenLLM Leaderboard Performance
|
160 |
+
|
161 |
+
| Models | Avg. | ARC | HellaSwag | MMLU | TruthfulQA | Winogrande | GSM8k |
|
162 |
+
|-----------------------------------------|------|-------|-----------|------|------------|------------|-------|
|
163 |
+
| google/gemma-2b | 46.37| 48.38 | 71.77 | 41.77| 33.08 | 66.77 | 16.91 |
|
164 |
+
| google/gemma-2b-it | 42.75| 43.94 | 62.70 | 37.65| 45.82 | 60.93 | 5.46 |
|
165 |
+
| wandb/gemma-2b-zephyr-sft | 47.18| 49.74 | 72.38 | 41.37| 34.42 | 66.93 | 18.27 |
|
166 |
+
| wandb/gemma-2b-zephyr-dpo | 46.92| 49.66 | 72.23 | 41.13| 34.47 | 66.54 | 17.51 |
|
167 |
+
| **Columbia-NLP/gemma-2b-zephyr-sft** | 48.75| 51.80 | 72.63 | 42.20| 41.96 | 63.85 | 20.09 |
|
168 |
+
| Columbia-NLP/gemma-2b-zephyr-dpo | 49.14| 52.22 | 73.11 | 42.55| 42.64 | 64.40 | 19.94 |
|
169 |
+
|
170 |
+
|
171 |
+
## MT-Bench
|
172 |
+
|
173 |
+
GPT-4-0125-preview as Judge
|
174 |
+
|
175 |
+
| Model | Total | Coding | Extraction | Humanities | Math | Reasoning | Roleplay | STEM | Writing |
|
176 |
+
|------------------------------------------|-------|--------|------------|------------|------|-----------|----------|------|---------|
|
177 |
+
| google/gemma-2b-it | 4.71 | 2.95 | 4.35 | 6.15 | 2.90 | 3.50 | 5.60 | 5.50 | 6.70 |
|
178 |
+
| wandb/gemma-2b-zephyr-sft | 4.03 | 3.10 | 3.15 | 5.00 | 2.70 | 2.65 | 5.10 | 4.80 | 5.75 |
|
179 |
+
| wandb/gemma-2b-zephyr-dpo | 4.06 | 2.80 | 2.90 | 5.55 | 2.65 | 2.70 | 5.20 | 4.80 | 5.85 |
|
180 |
+
| **Columbia-NLP/gemma-2b-zephyr-sft** | 4.34 | 3.10 | 3.70 | 6.25 | 2.65 | 2.70 | 5.55 | 5.25 | 5.50 |
|
181 |
+
| Columbia-NLP/gemma-2b-zephyr-dpo | 4.75 | 3.50 | 4.05 | 6.75 | 3.30 | 3.70 | 5.85 | 5.40 | 5.53 |
|
182 |
+
|
183 |
+
|
184 |
+
|
185 |
+
|
186 |
+
|