Training procedure
The following bitsandbytes
quantization config was used during training:
- quant_method: bitsandbytes
- _load_in_8bit: False
- _load_in_4bit: True
- llm_int8_threshold: 6.0
- llm_int8_skip_modules: None
- llm_int8_enable_fp32_cpu_offload: False
- llm_int8_has_fp16_weight: False
- bnb_4bit_quant_type: nf4
- bnb_4bit_use_double_quant: True
- bnb_4bit_compute_dtype: float16
- bnb_4bit_quant_storage: uint8
- load_in_4bit: True
- load_in_8bit: False
Framework versions
- PEFT 0.5.0
Open Portuguese LLM Leaderboard Evaluation Results
Detailed results can be found here and on the ๐ Open Portuguese LLM Leaderboard
Metric | Value |
---|---|
Average | 32.3 |
ENEM Challenge (No Images) | 24.14 |
BLUEX (No Images) | 20.31 |
OAB Exams | 25.56 |
Assin2 RTE | 69.75 |
Assin2 STS | 4.16 |
FaQuAD NLI | 52.63 |
HateBR Binary | 33.33 |
PT Hate Speech Binary | 41.65 |
tweetSentBR | 19.15 |
- Downloads last month
- 2
Unable to determine this modelโs pipeline type. Check the
docs
.
Space using recogna-nlp/gembode-2b-base-ultraalpaca-qlora 1
Evaluation results
- accuracy on ENEM Challenge (No Images)Open Portuguese LLM Leaderboard24.140
- accuracy on BLUEX (No Images)Open Portuguese LLM Leaderboard20.310
- accuracy on OAB ExamsOpen Portuguese LLM Leaderboard25.560
- f1-macro on Assin2 RTEtest set Open Portuguese LLM Leaderboard69.750
- pearson on Assin2 STStest set Open Portuguese LLM Leaderboard4.160
- f1-macro on FaQuAD NLItest set Open Portuguese LLM Leaderboard52.630
- f1-macro on HateBR Binarytest set Open Portuguese LLM Leaderboard33.330
- f1-macro on PT Hate Speech Binarytest set Open Portuguese LLM Leaderboard41.650
- f1-macro on tweetSentBRtest set Open Portuguese LLM Leaderboard19.150