RichardErkhov/ZhangShenao_-_SELM-Llama-3-8B-Instruct-iter-3-gguf

Quantization made by Richard Erkhov.

Github

Discord

Request more models

SELM-Llama-3-8B-Instruct-iter-3 - GGUF

• Model creator: https://huggingface.co/ZhangShenao/
• Original model: https://huggingface.co/ZhangShenao/SELM-Llama-3-8B-Instruct-iter-3/

Name	Quant method	Size
SELM-Llama-3-8B-Instruct-iter-3.Q2_K.gguf	Q2_K	2.96GB
SELM-Llama-3-8B-Instruct-iter-3.IQ3_XS.gguf	IQ3_XS	3.28GB
SELM-Llama-3-8B-Instruct-iter-3.IQ3_S.gguf	IQ3_S	3.43GB
SELM-Llama-3-8B-Instruct-iter-3.Q3_K_S.gguf	Q3_K_S	3.41GB
SELM-Llama-3-8B-Instruct-iter-3.IQ3_M.gguf	IQ3_M	3.52GB
SELM-Llama-3-8B-Instruct-iter-3.Q3_K.gguf	Q3_K	3.74GB
SELM-Llama-3-8B-Instruct-iter-3.Q3_K_M.gguf	Q3_K_M	3.74GB
SELM-Llama-3-8B-Instruct-iter-3.Q3_K_L.gguf	Q3_K_L	4.03GB
SELM-Llama-3-8B-Instruct-iter-3.IQ4_XS.gguf	IQ4_XS	4.18GB
SELM-Llama-3-8B-Instruct-iter-3.Q4_0.gguf	Q4_0	4.34GB
SELM-Llama-3-8B-Instruct-iter-3.IQ4_NL.gguf	IQ4_NL	4.38GB
SELM-Llama-3-8B-Instruct-iter-3.Q4_K_S.gguf	Q4_K_S	4.37GB
SELM-Llama-3-8B-Instruct-iter-3.Q4_K.gguf	Q4_K	4.58GB
SELM-Llama-3-8B-Instruct-iter-3.Q4_K_M.gguf	Q4_K_M	4.58GB
SELM-Llama-3-8B-Instruct-iter-3.Q4_1.gguf	Q4_1	4.78GB
SELM-Llama-3-8B-Instruct-iter-3.Q5_0.gguf	Q5_0	5.21GB
SELM-Llama-3-8B-Instruct-iter-3.Q5_K_S.gguf	Q5_K_S	5.21GB
SELM-Llama-3-8B-Instruct-iter-3.Q5_K.gguf	Q5_K	5.34GB
SELM-Llama-3-8B-Instruct-iter-3.Q5_K_M.gguf	Q5_K_M	5.34GB
SELM-Llama-3-8B-Instruct-iter-3.Q5_1.gguf	Q5_1	5.65GB
SELM-Llama-3-8B-Instruct-iter-3.Q6_K.gguf	Q6_K	6.14GB
SELM-Llama-3-8B-Instruct-iter-3.Q8_0.gguf	Q8_0	7.95GB

Original model description:

license: mit base_model: ZhangShenao/SELM-Llama-3-8B-Instruct-iter-2 tags: - alignment-handbook - dpo - trl - selm datasets: - HuggingFaceH4/ultrafeedback_binarized model-index: - name: SELM-Llama-3-8B-Instruct-iter-3 results: []

Self-Exploring Language Models: Active Preference Elicitation for Online Alignment.

SELM-Llama-3-8B-Instruct-iter-3

This model is a fine-tuned version of ZhangShenao/SELM-Llama-3-8B-Instruct-iter-2 using synthetic data based on on the HuggingFaceH4/ultrafeedback_binarized dataset.

Model description

• Model type: A 8B parameter Llama3-instruct-based Self-Exploring Language Models (SELM).
• License: MIT

Results

	AlpacaEval 2.0 (LC WR)	MT-Bench (Average)
SELM-Llama-3-8B-Instruct-iter-3	33.47	8.29
SELM-Llama-3-8B-Instruct-iter-2	35.65	8.09
SELM-Llama-3-8B-Instruct-iter-1	32.02	7.92
Meta-Llama-3-8B-Instruct	24.31	7.93

Our model also ranks highly on WildBench! 🔥

Training hyperparameters

The following hyperparameters were used during training:

• alpha: 0.0001
• beta: 0.01
• train_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 4
• total_train_batch_size: 128
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• num_epochs: 1

Framework versions

• Transformers 4.40.2
• Pytorch 2.1.2+cu121
• Datasets 2.14.6
• Tokenizers 0.19.1