license |
---|
apache-2.0 |
This model is a finetuning of other models based on mistralai/Mistral-7B-v0.1.
The model has been generated from the merging of the models viethq188/LeoScorpius-7B-Chat-DPO and GreenNode/GreenNodeLM-7B-v1olet and a later finetuning with a Platypus dataset garage-bAInd/Open-Platypus.
- Developed by: Ignos
- Model type: Mistral
- License: Apache-2.0
The model aims to have good overall comparative results on HuggingFace metrics, improving reasoning.
The same bias, risks and limitations from base models.
- Training with QLoRA approach and merging with base model.
- Huggingface evaluation pending
- Models based on Mistral Architecture
- Training on RunPod
- 4 x Nvidia RTX 4090
- 64 vCPU 503 GB RAM
- Mergekit (main)
- Axolotl 0.3.0
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: bfloat16
- PEFT 0.6.0