munish0838
commited on
Commit
•
b0f6a0d
1
Parent(s):
9d8872d
Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,238 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
language:
|
3 |
+
- pt
|
4 |
+
license: apache-2.0
|
5 |
+
base_model: botbot-ai/CabraMistral-v3-7b-32k
|
6 |
+
model-index:
|
7 |
+
- name: CabraMistral-v3-7b-32k
|
8 |
+
results:
|
9 |
+
- task:
|
10 |
+
type: text-generation
|
11 |
+
name: Text Generation
|
12 |
+
dataset:
|
13 |
+
name: ENEM Challenge (No Images)
|
14 |
+
type: eduagarcia/enem_challenge
|
15 |
+
split: train
|
16 |
+
args:
|
17 |
+
num_few_shot: 3
|
18 |
+
metrics:
|
19 |
+
- type: acc
|
20 |
+
value: 58.64
|
21 |
+
name: accuracy
|
22 |
+
source:
|
23 |
+
url: >-
|
24 |
+
https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=botbot-ai/CabraMistral-v3-7b-32k
|
25 |
+
name: Open Portuguese LLM Leaderboard
|
26 |
+
- task:
|
27 |
+
type: text-generation
|
28 |
+
name: Text Generation
|
29 |
+
dataset:
|
30 |
+
name: BLUEX (No Images)
|
31 |
+
type: eduagarcia-temp/BLUEX_without_images
|
32 |
+
split: train
|
33 |
+
args:
|
34 |
+
num_few_shot: 3
|
35 |
+
metrics:
|
36 |
+
- type: acc
|
37 |
+
value: 45.62
|
38 |
+
name: accuracy
|
39 |
+
source:
|
40 |
+
url: >-
|
41 |
+
https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=botbot-ai/CabraMistral-v3-7b-32k
|
42 |
+
name: Open Portuguese LLM Leaderboard
|
43 |
+
- task:
|
44 |
+
type: text-generation
|
45 |
+
name: Text Generation
|
46 |
+
dataset:
|
47 |
+
name: OAB Exams
|
48 |
+
type: eduagarcia/oab_exams
|
49 |
+
split: train
|
50 |
+
args:
|
51 |
+
num_few_shot: 3
|
52 |
+
metrics:
|
53 |
+
- type: acc
|
54 |
+
value: 41.46
|
55 |
+
name: accuracy
|
56 |
+
source:
|
57 |
+
url: >-
|
58 |
+
https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=botbot-ai/CabraMistral-v3-7b-32k
|
59 |
+
name: Open Portuguese LLM Leaderboard
|
60 |
+
- task:
|
61 |
+
type: text-generation
|
62 |
+
name: Text Generation
|
63 |
+
dataset:
|
64 |
+
name: Assin2 RTE
|
65 |
+
type: assin2
|
66 |
+
split: test
|
67 |
+
args:
|
68 |
+
num_few_shot: 15
|
69 |
+
metrics:
|
70 |
+
- type: f1_macro
|
71 |
+
value: 86.14
|
72 |
+
name: f1-macro
|
73 |
+
source:
|
74 |
+
url: >-
|
75 |
+
https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=botbot-ai/CabraMistral-v3-7b-32k
|
76 |
+
name: Open Portuguese LLM Leaderboard
|
77 |
+
- task:
|
78 |
+
type: text-generation
|
79 |
+
name: Text Generation
|
80 |
+
dataset:
|
81 |
+
name: Assin2 STS
|
82 |
+
type: eduagarcia/portuguese_benchmark
|
83 |
+
split: test
|
84 |
+
args:
|
85 |
+
num_few_shot: 15
|
86 |
+
metrics:
|
87 |
+
- type: pearson
|
88 |
+
value: 68.06
|
89 |
+
name: pearson
|
90 |
+
source:
|
91 |
+
url: >-
|
92 |
+
https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=botbot-ai/CabraMistral-v3-7b-32k
|
93 |
+
name: Open Portuguese LLM Leaderboard
|
94 |
+
- task:
|
95 |
+
type: text-generation
|
96 |
+
name: Text Generation
|
97 |
+
dataset:
|
98 |
+
name: FaQuAD NLI
|
99 |
+
type: ruanchaves/faquad-nli
|
100 |
+
split: test
|
101 |
+
args:
|
102 |
+
num_few_shot: 15
|
103 |
+
metrics:
|
104 |
+
- type: f1_macro
|
105 |
+
value: 47.46
|
106 |
+
name: f1-macro
|
107 |
+
source:
|
108 |
+
url: >-
|
109 |
+
https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=botbot-ai/CabraMistral-v3-7b-32k
|
110 |
+
name: Open Portuguese LLM Leaderboard
|
111 |
+
- task:
|
112 |
+
type: text-generation
|
113 |
+
name: Text Generation
|
114 |
+
dataset:
|
115 |
+
name: HateBR Binary
|
116 |
+
type: ruanchaves/hatebr
|
117 |
+
split: test
|
118 |
+
args:
|
119 |
+
num_few_shot: 25
|
120 |
+
metrics:
|
121 |
+
- type: f1_macro
|
122 |
+
value: 70.46
|
123 |
+
name: f1-macro
|
124 |
+
source:
|
125 |
+
url: >-
|
126 |
+
https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=botbot-ai/CabraMistral-v3-7b-32k
|
127 |
+
name: Open Portuguese LLM Leaderboard
|
128 |
+
- task:
|
129 |
+
type: text-generation
|
130 |
+
name: Text Generation
|
131 |
+
dataset:
|
132 |
+
name: PT Hate Speech Binary
|
133 |
+
type: hate_speech_portuguese
|
134 |
+
split: test
|
135 |
+
args:
|
136 |
+
num_few_shot: 25
|
137 |
+
metrics:
|
138 |
+
- type: f1_macro
|
139 |
+
value: 62.39
|
140 |
+
name: f1-macro
|
141 |
+
source:
|
142 |
+
url: >-
|
143 |
+
https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=botbot-ai/CabraMistral-v3-7b-32k
|
144 |
+
name: Open Portuguese LLM Leaderboard
|
145 |
+
- task:
|
146 |
+
type: text-generation
|
147 |
+
name: Text Generation
|
148 |
+
dataset:
|
149 |
+
name: tweetSentBR
|
150 |
+
type: eduagarcia/tweetsentbr_fewshot
|
151 |
+
split: test
|
152 |
+
args:
|
153 |
+
num_few_shot: 25
|
154 |
+
metrics:
|
155 |
+
- type: f1_macro
|
156 |
+
value: 65.71
|
157 |
+
name: f1-macro
|
158 |
+
source:
|
159 |
+
url: >-
|
160 |
+
https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=botbot-ai/CabraMistral-v3-7b-32k
|
161 |
+
name: Open Portuguese LLM Leaderboard
|
162 |
+
pipeline_tag: text-generation
|
163 |
+
---
|
164 |
+
|
165 |
+
# QuantFactory/CabraMistral-v3-7b-32k-GGUF
|
166 |
+
This is quantized version of [botbot-ai/CabraMistral-v3-7b-32k](https://huggingface.co/botbot-ai/CabraMistral-v3-7b-32k) created using llama.cpp
|
167 |
+
|
168 |
+
# Model Description
|
169 |
+
<img src="https://uploads-ssl.webflow.com/65f77c0240ae1c68f8192771/660b1a4d574293d8a1ce48ca_cabra1.png" width="400" height="400">
|
170 |
+
|
171 |
+
Esse modelo é um finetune do [Mistral 7b Instruct 0.3](https://huggingface.co/mistralai/mistral-7b-instruct-v0.3) com o dataset BotBot Cabra 10k. Esse modelo é optimizado para português.
|
172 |
+
|
173 |
+
**Conheça os nossos outros modelos: [Cabra](https://huggingface.co/collections/botbot-ai/models-6604c2069ceef04f834ba99b).**
|
174 |
+
|
175 |
+
## Detalhes do Modelo
|
176 |
+
|
177 |
+
### Modelo: Mistral 7b Instruct 0.3
|
178 |
+
|
179 |
+
Mistral-7B-v0.3 é um modelo de transformador, com as seguintes escolhas arquitetônicas:
|
180 |
+
|
181 |
+
- Grouped-Query Attention
|
182 |
+
- Sliding-Window Attention
|
183 |
+
- Byte-fallback BPE tokenizer
|
184 |
+
|
185 |
+
### dataset: Cabra 10k
|
186 |
+
|
187 |
+
Dataset interno para finetuning. Vamos lançar em breve.
|
188 |
+
|
189 |
+
### Exemplo
|
190 |
+
|
191 |
+
```
|
192 |
+
<s> [INST] who is Elon Musk? [/INST]Elon Musk é um empreendedor, inventor e capitalista americano. Ele é o fundador, CEO e CTO da SpaceX, CEO da Neuralink e fundador do The Boring Company. Musk também é o proprietário do Twitter.</s>
|
193 |
+
```
|
194 |
+
|
195 |
+
### Paramentros de trainamento
|
196 |
+
|
197 |
+
```
|
198 |
+
- learning_rate: 1e-05
|
199 |
+
- train_batch_size: 4
|
200 |
+
- eval_batch_size: 4
|
201 |
+
- seed: 42
|
202 |
+
- distributed_type: multi-GPU
|
203 |
+
- num_devices: 2
|
204 |
+
- gradient_accumulation_steps: 8
|
205 |
+
- total_train_batch_size: 64
|
206 |
+
- total_eval_batch_size: 8
|
207 |
+
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
208 |
+
- lr_scheduler_type: cosine
|
209 |
+
- lr_scheduler_warmup_ratio: 0.01
|
210 |
+
- num_epochs: 3
|
211 |
+
```
|
212 |
+
|
213 |
+
### Framework
|
214 |
+
|
215 |
+
- Transformers 4.39.0.dev0
|
216 |
+
- Pytorch 2.1.2+cu118
|
217 |
+
- Datasets 2.14.6
|
218 |
+
- Tokenizers 0.15.2
|
219 |
+
|
220 |
+
### Evals
|
221 |
+
|
222 |
+
|
223 |
+
# Open Portuguese LLM Leaderboard Evaluation Results
|
224 |
+
|
225 |
+
Detailed results can be found [here](https://huggingface.co/datasets/eduagarcia-temp/llm_pt_leaderboard_raw_results/tree/main/botbot-ai/CabraMistral-v3-7b-32k) and on the [🚀 Open Portuguese LLM Leaderboard](https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard)
|
226 |
+
|
227 |
+
| Metric | Value |
|
228 |
+
|--------------------------|---------|
|
229 |
+
|Average |**60.66**|
|
230 |
+
|ENEM Challenge (No Images)| 58.64|
|
231 |
+
|BLUEX (No Images) | 45.62|
|
232 |
+
|OAB Exams | 41.46|
|
233 |
+
|Assin2 RTE | 86.14|
|
234 |
+
|Assin2 STS | 68.06|
|
235 |
+
|FaQuAD NLI | 47.46|
|
236 |
+
|HateBR Binary | 70.46|
|
237 |
+
|PT Hate Speech Binary | 62.39|
|
238 |
+
|tweetSentBR | 65.71|
|