JGLUE Score
I evaluated this model using the following JGLUE tasks. Here are the scores:
| Task | stablelm-base-alpha-7b | This Model | stablelm-instruct-alpha-7b |
|---|---|---|---|
| JCOMMONSENSEQA(acc) | 33.42 | 79.17 | 82.22 |
| JNLI(acc) | 43.34 | 47.82 | 52.05 |
| MARC_JA(acc) | 96.73 | 88.14 | 82.88 |
| JSQUAD(exact_match) | 70.62 | 29.85 | 63.26 |
| Average | 61.03 | 61.25 | 70.10 |
- Note: Use v0.3 prompt template
- The JGLUE scores were measured using the following script: Stability-AI/lm-evaluation-harness
- The JGLUE scores of Model "stablelm-base-alpha-7b" and "stablelm-instruct-alpha-7b" were referenced from Github above.
Training procedure
The following bitsandbytes quantization config was used during training:
- 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: False
- bnb_4bit_compute_dtype: float16
Framework versions
- PEFT 0.4.0
