phpaiola commited on
Commit
aea7cd1
1 Parent(s): ea30834

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +11 -17
README.md CHANGED
@@ -148,27 +148,21 @@ model-index:
148
  url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=recogna-nlp/gembode-7b
149
  name: Open Portuguese LLM Leaderboard
150
  ---
151
- ## Training procedure
152
 
 
153
 
154
- The following `bitsandbytes` quantization config was used during training:
155
- - quant_method: bitsandbytes
156
- - _load_in_8bit: False
157
- - _load_in_4bit: True
158
- - llm_int8_threshold: 6.0
159
- - llm_int8_skip_modules: None
160
- - llm_int8_enable_fp32_cpu_offload: False
161
- - llm_int8_has_fp16_weight: False
162
- - bnb_4bit_quant_type: nf4
163
- - bnb_4bit_use_double_quant: True
164
- - bnb_4bit_compute_dtype: float16
165
- - bnb_4bit_quant_storage: uint8
166
- - load_in_4bit: True
167
- - load_in_8bit: False
168
- ### Framework versions
169
 
 
170
 
171
- - PEFT 0.5.0
 
 
 
 
172
 
173
 
174
  # Open Portuguese LLM Leaderboard Evaluation Results
 
148
  url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=recogna-nlp/gembode-7b
149
  name: Open Portuguese LLM Leaderboard
150
  ---
 
151
 
152
+ # gembode-7b
153
 
154
+ <!--- PROJECT LOGO -->
155
+ <p align="center">
156
+ <img src="https://huggingface.co/recogna-nlp/GemBode-2b-it/resolve/main/gembode.jpg" alt="Phi-Bode Logo" width="400" style="margin-left:'auto' margin-right:'auto' display:'block'"/>
157
+ </p>
 
 
 
 
 
 
 
 
 
 
 
158
 
159
+ GemmBode é um modelo de linguagem ajustado para o idioma português, desenvolvido a partir do modelo Gemma-7b fornecido pela [Google](https://huggingface.co/google/gemma-7b).
160
 
161
+ ## Características Principais
162
+
163
+ - **Modelo Base:** Gemma-7b, criado pela Google, com 7 bilhões de parâmetros.
164
+ - **Dataset para Fine-tuning:** UltraAlpaca
165
+ - **Treinamento:** O treinamento foi realizado a partir do fine-tuning, com QLoRA do gemma-7b.
166
 
167
 
168
  # Open Portuguese LLM Leaderboard Evaluation Results