Quantization made by Richard Erkhov.

gemma-2-Ifable-9B - GGUF

Model creator: https://huggingface.co./ifable/
Original model: https://huggingface.co./ifable/gemma-2-Ifable-9B/

Name	Quant method	Size
gemma-2-Ifable-9B.Q2_K.gguf	Q2_K	3.54GB
gemma-2-Ifable-9B.IQ3_XS.gguf	IQ3_XS	3.86GB
gemma-2-Ifable-9B.IQ3_S.gguf	IQ3_S	4.04GB
gemma-2-Ifable-9B.Q3_K_S.gguf	Q3_K_S	4.04GB
gemma-2-Ifable-9B.IQ3_M.gguf	IQ3_M	4.19GB
gemma-2-Ifable-9B.Q3_K.gguf	Q3_K	4.43GB
gemma-2-Ifable-9B.Q3_K_M.gguf	Q3_K_M	4.43GB
gemma-2-Ifable-9B.Q3_K_L.gguf	Q3_K_L	4.78GB
gemma-2-Ifable-9B.IQ4_XS.gguf	IQ4_XS	4.86GB
gemma-2-Ifable-9B.Q4_0.gguf	Q4_0	5.07GB
gemma-2-Ifable-9B.IQ4_NL.gguf	IQ4_NL	5.1GB
gemma-2-Ifable-9B.Q4_K_S.gguf	Q4_K_S	5.1GB
gemma-2-Ifable-9B.Q4_K.gguf	Q4_K	5.37GB
gemma-2-Ifable-9B.Q4_K_M.gguf	Q4_K_M	5.37GB
gemma-2-Ifable-9B.Q4_1.gguf	Q4_1	5.55GB
gemma-2-Ifable-9B.Q5_0.gguf	Q5_0	6.04GB
gemma-2-Ifable-9B.Q5_K_S.gguf	Q5_K_S	6.04GB
gemma-2-Ifable-9B.Q5_K.gguf	Q5_K	6.19GB
gemma-2-Ifable-9B.Q5_K_M.gguf	Q5_K_M	6.19GB
gemma-2-Ifable-9B.Q5_1.gguf	Q5_1	6.52GB
gemma-2-Ifable-9B.Q6_K.gguf	Q6_K	7.07GB
gemma-2-Ifable-9B.Q8_0.gguf	Q8_0	9.15GB

Original model description:

license: gemma library_name: transformers datasets: - jondurbin/gutenberg-dpo-v0.1

ifable/gemma-2-Ifable-9B

This model ranked first on the Creative Writing Benchmark (https://eqbench.com/creative_writing.html) on September 10, 2024

Training and evaluation data

Gutenberg: https://huggingface.co./datasets/jondurbin/gutenberg-dpo-v0.1
Carefully curated proprietary creative writing dataset

Training procedure

Training method: SimPO (GitHub - princeton-nlp/SimPO: SimPO: Simple Preference Optimization with a Reference-Free Reward)

It achieves the following results on the evaluation set:

Loss: 1.0163
Rewards/chosen: -21.6822
Rewards/rejected: -47.8754
Rewards/accuracies: 0.9167
Rewards/margins: 26.1931
Logps/rejected: -4.7875
Logps/chosen: -2.1682
Logits/rejected: -17.0475
Logits/chosen: -12.0041

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 8e-07
train_batch_size: 1
eval_batch_size: 1
seed: 42
distributed_type: multi-GPU
num_devices: 8
gradient_accumulation_steps: 16
total_train_batch_size: 128
total_eval_batch_size: 8
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
lr_scheduler_warmup_ratio: 0.1
num_epochs: 1.0

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen	Sft Loss
1.4444	0.9807	35	1.0163	-21.6822	-47.8754	0.9167	26.1931	-4.7875	-2.1682	-17.0475	-12.0041	0.0184

Framework versions

Transformers 4.43.4
Pytorch 2.3.0a0+ebedce2
Datasets 2.20.0
Tokenizers 0.19.1

We are looking for product manager and operations managers to build applications through our model, and also open for business cooperation, and also AI engineer to join us, contact with : [email protected]