aashish1904 commited on
Commit
49a804f
1 Parent(s): 64e89c0

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +303 -0
README.md ADDED
@@ -0,0 +1,303 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+ ---
3
+
4
+ language:
5
+ - pt
6
+ license: apache-2.0
7
+ library_name: transformers
8
+ tags:
9
+ - portugues
10
+ - portuguese
11
+ - QA
12
+ - instruct
13
+ - phi
14
+ base_model: microsoft/Phi-3-mini-4k-instruct
15
+ datasets:
16
+ - rhaymison/superset
17
+ pipeline_tag: text-generation
18
+ model-index:
19
+ - name: phi-3-portuguese-tom-cat-4k-instruct
20
+ results:
21
+ - task:
22
+ type: text-generation
23
+ name: Text Generation
24
+ dataset:
25
+ name: ENEM Challenge (No Images)
26
+ type: eduagarcia/enem_challenge
27
+ split: train
28
+ args:
29
+ num_few_shot: 3
30
+ metrics:
31
+ - type: acc
32
+ value: 61.58
33
+ name: accuracy
34
+ source:
35
+ url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=rhaymison/phi-3-portuguese-tom-cat-4k-instruct
36
+ name: Open Portuguese LLM Leaderboard
37
+ - task:
38
+ type: text-generation
39
+ name: Text Generation
40
+ dataset:
41
+ name: BLUEX (No Images)
42
+ type: eduagarcia-temp/BLUEX_without_images
43
+ split: train
44
+ args:
45
+ num_few_shot: 3
46
+ metrics:
47
+ - type: acc
48
+ value: 50.63
49
+ name: accuracy
50
+ source:
51
+ url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=rhaymison/phi-3-portuguese-tom-cat-4k-instruct
52
+ name: Open Portuguese LLM Leaderboard
53
+ - task:
54
+ type: text-generation
55
+ name: Text Generation
56
+ dataset:
57
+ name: OAB Exams
58
+ type: eduagarcia/oab_exams
59
+ split: train
60
+ args:
61
+ num_few_shot: 3
62
+ metrics:
63
+ - type: acc
64
+ value: 43.69
65
+ name: accuracy
66
+ source:
67
+ url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=rhaymison/phi-3-portuguese-tom-cat-4k-instruct
68
+ name: Open Portuguese LLM Leaderboard
69
+ - task:
70
+ type: text-generation
71
+ name: Text Generation
72
+ dataset:
73
+ name: Assin2 RTE
74
+ type: assin2
75
+ split: test
76
+ args:
77
+ num_few_shot: 15
78
+ metrics:
79
+ - type: f1_macro
80
+ value: 91.54
81
+ name: f1-macro
82
+ source:
83
+ url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=rhaymison/phi-3-portuguese-tom-cat-4k-instruct
84
+ name: Open Portuguese LLM Leaderboard
85
+ - task:
86
+ type: text-generation
87
+ name: Text Generation
88
+ dataset:
89
+ name: Assin2 STS
90
+ type: eduagarcia/portuguese_benchmark
91
+ split: test
92
+ args:
93
+ num_few_shot: 15
94
+ metrics:
95
+ - type: pearson
96
+ value: 75.27
97
+ name: pearson
98
+ source:
99
+ url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=rhaymison/phi-3-portuguese-tom-cat-4k-instruct
100
+ name: Open Portuguese LLM Leaderboard
101
+ - task:
102
+ type: text-generation
103
+ name: Text Generation
104
+ dataset:
105
+ name: FaQuAD NLI
106
+ type: ruanchaves/faquad-nli
107
+ split: test
108
+ args:
109
+ num_few_shot: 15
110
+ metrics:
111
+ - type: f1_macro
112
+ value: 47.46
113
+ name: f1-macro
114
+ source:
115
+ url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=rhaymison/phi-3-portuguese-tom-cat-4k-instruct
116
+ name: Open Portuguese LLM Leaderboard
117
+ - task:
118
+ type: text-generation
119
+ name: Text Generation
120
+ dataset:
121
+ name: HateBR Binary
122
+ type: ruanchaves/hatebr
123
+ split: test
124
+ args:
125
+ num_few_shot: 25
126
+ metrics:
127
+ - type: f1_macro
128
+ value: 83.01
129
+ name: f1-macro
130
+ source:
131
+ url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=rhaymison/phi-3-portuguese-tom-cat-4k-instruct
132
+ name: Open Portuguese LLM Leaderboard
133
+ - task:
134
+ type: text-generation
135
+ name: Text Generation
136
+ dataset:
137
+ name: PT Hate Speech Binary
138
+ type: hate_speech_portuguese
139
+ split: test
140
+ args:
141
+ num_few_shot: 25
142
+ metrics:
143
+ - type: f1_macro
144
+ value: 70.19
145
+ name: f1-macro
146
+ source:
147
+ url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=rhaymison/phi-3-portuguese-tom-cat-4k-instruct
148
+ name: Open Portuguese LLM Leaderboard
149
+ - task:
150
+ type: text-generation
151
+ name: Text Generation
152
+ dataset:
153
+ name: tweetSentBR
154
+ type: eduagarcia/tweetsentbr_fewshot
155
+ split: test
156
+ args:
157
+ num_few_shot: 25
158
+ metrics:
159
+ - type: f1_macro
160
+ value: 57.78
161
+ name: f1-macro
162
+ source:
163
+ url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=rhaymison/phi-3-portuguese-tom-cat-4k-instruct
164
+ name: Open Portuguese LLM Leaderboard
165
+
166
+ ---
167
+
168
+ [![QuantFactory Banner](https://lh7-rt.googleusercontent.com/docsz/AD_4nXeiuCm7c8lEwEJuRey9kiVZsRn2W-b4pWlu3-X534V3YmVuVc2ZL-NXg2RkzSOOS2JXGHutDuyyNAUtdJI65jGTo8jT9Y99tMi4H4MqL44Uc5QKG77B0d6-JfIkZHFaUA71-RtjyYZWVIhqsNZcx8-OMaA?key=xt3VSDoCbmTY7o-cwwOFwQ)](https://hf.co/QuantFactory)
169
+
170
+
171
+ # QuantFactory/phi-3-portuguese-tom-cat-4k-instruct-GGUF
172
+ This is quantized version of [rhaymison/phi-3-portuguese-tom-cat-4k-instruct](https://huggingface.co/rhaymison/phi-3-portuguese-tom-cat-4k-instruct) created using llama.cpp
173
+
174
+ # Original Model Card
175
+
176
+
177
+ # Phi-3-portuguese-tom-cat-4k-instruct
178
+
179
+ <p align="center">
180
+ <img src="https://raw.githubusercontent.com/rhaymisonbetini/huggphotos/main/tom-cat.webp" width="50%" style="margin-left:'auto' margin-right:'auto' display:'block'"/>
181
+ </p>
182
+
183
+
184
+ This model was trained with a superset of 300,000 instructions in Portuguese.
185
+ The model comes to help fill the gap in models in Portuguese. Tuned from the microsoft/Phi-3-mini-4k.
186
+
187
+ If you are looking for enhanced compatibility, the Luana model also has a GGUF family that can be run with LlamaCpp.
188
+ You can explore the GGUF models starting with the one below:
189
+
190
+ - [rhaymison/phi-3-portuguese-tom-cat-4k-instruct-q8-gguf](https://huggingface.co/rhaymison/phi-3-portuguese-tom-cat-4k-instruct-q8-gguf)
191
+ - [rhaymison/phi-3-portuguese-tom-cat-4k-instruct-f16-gguf](https://huggingface.co/rhaymison/phi-3-portuguese-tom-cat-4k-instruct-f16-gguf)
192
+
193
+ Explore this and other models to find the best fit for your needs!
194
+
195
+ # How to use
196
+
197
+ ### FULL MODEL : A100
198
+ ### HALF MODEL: L4
199
+ ### 8bit or 4bit : T4 or V100
200
+
201
+ You can use the model in its normal form up to 4-bit quantization. Below we will use both approaches.
202
+ Remember that verbs are important in your prompt. Tell your model how to act or behave so that you can guide them along the path of their response.
203
+ Important points like these help models (even smaller models like 4b) to perform much better.
204
+
205
+ ```python
206
+ !pip install -q -U transformers
207
+ !pip install -q -U accelerate
208
+ !pip install -q -U bitsandbytes
209
+
210
+ from transformers import AutoModelForCausalLM, AutoTokenizer, TextStreamer
211
+ model = AutoModelForCausalLM.from_pretrained("rhaymison/phi-3-portuguese-tom-cat-4k-instruct", device_map= {"": 0})
212
+ tokenizer = AutoTokenizer.from_pretrained("rhaymison/phi-3-portuguese-tom-cat-4k-instruct")
213
+ model.eval()
214
+
215
+ ```
216
+
217
+ You can use with Pipeline.
218
+ ```python
219
+
220
+ from transformers import pipeline
221
+ pipe = pipeline("text-generation",
222
+ model=model,
223
+ tokenizer=tokenizer,
224
+ do_sample=True,
225
+ max_new_tokens=512,
226
+ num_beams=2,
227
+ temperature=0.3,
228
+ top_k=50,
229
+ top_p=0.95,
230
+ early_stopping=True,
231
+ pad_token_id=tokenizer.eos_token_id,
232
+ )
233
+
234
+
235
+ def format_template(question:str):
236
+ system_prompt = "Abaixo está uma instrução que descreve uma tarefa, juntamente com uma entrada que fornece mais contexto. Escreva uma resposta que complete adequadamente o pedido."
237
+ return f"""<s><|system|>
238
+ { system_prompt }
239
+ <|user|>
240
+ { question }
241
+ <|assistant|>
242
+ """
243
+
244
+ question = format_template("E possivel ir de Carro dos Estados unidos ate o japão")
245
+ pipe(question)
246
+ ```
247
+
248
+ If you are having a memory problem such as "CUDA Out of memory", you should use 4-bit or 8-bit quantization.
249
+ For the complete model in colab you will need the A100.
250
+ If you want to use 4bits or 8bits, T4 or L4 will already solve the problem.
251
+
252
+ # 4bits example
253
+
254
+ ```python
255
+ from transformers import BitsAndBytesConfig
256
+ import torch
257
+ nb_4bit_config = BitsAndBytesConfig(
258
+ load_in_4bit=True,
259
+ bnb_4bit_quant_type="nf4",
260
+ bnb_4bit_compute_dtype=torch.bfloat16,
261
+ bnb_4bit_use_double_quant=True
262
+ )
263
+
264
+ model = AutoModelForCausalLM.from_pretrained(
265
+ base_model,
266
+ quantization_config=bnb_config,
267
+ device_map={"": 0}
268
+ )
269
+
270
+ ```
271
+
272
+ # Open Portuguese LLM Leaderboard Evaluation Results
273
+
274
+ Detailed results can be found [here](https://huggingface.co/datasets/eduagarcia-temp/llm_pt_leaderboard_raw_results/tree/main/rhaymison/phi-3-portuguese-tom-cat-4k-instruct) and on the [🚀 Open Portuguese LLM Leaderboard](https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard)
275
+
276
+ | Metric | Value |
277
+ |--------------------------|---------|
278
+ |Average |**64.57**|
279
+ |ENEM Challenge (No Images)| 61.58|
280
+ |BLUEX (No Images) | 50.63|
281
+ |OAB Exams | 43.69|
282
+ |Assin2 RTE | 91.54|
283
+ |Assin2 STS | 75.27|
284
+ |FaQuAD NLI | 47.46|
285
+ |HateBR Binary | 83.01|
286
+ |PT Hate Speech Binary | 70.19|
287
+ |tweetSentBR | 57.78|
288
+
289
+
290
+ ### Comments
291
+
292
+ Any idea, help or report will always be welcome.
293
+
294
+ email: rhaymisoncristian@gmail.com
295
+
296
+ <div style="display:flex; flex-direction:row; justify-content:left">
297
+ <a href="https://www.linkedin.com/in/rhaymison-cristian-betini-2b3016175/" target="_blank">
298
+ <img src="https://img.shields.io/badge/LinkedIn-0077B5?style=for-the-badge&logo=linkedin&logoColor=white">
299
+ </a>
300
+ <a href="https://github.com/rhaymisonbetini" target="_blank">
301
+ <img src="https://img.shields.io/badge/GitHub-100000?style=for-the-badge&logo=github&logoColor=white">
302
+ </a>
303
+