NeuralHermes-2.5-Mistral-7B
NeuralHermes-2.5-Mistral-7B: Unleash unparalleled AI capabilities with Mistral's 7B parameters for advanced natural language processing.
Last updated
NeuralHermes-2.5-Mistral-7B: Unleash unparalleled AI capabilities with Mistral's 7B parameters for advanced natural language processing.
Last updated
Developer Portal :
EXL2:
Update: NeuralHermes-2.5 became the best Hermes-based model on the Open LLM leaderboard and one of the very best 7b models. 🎉
Results are improved on every benchmark: AGIEval (from 43.07% to 43.62%), GPT4All (from 73.12% to 73.25%), and TruthfulQA.
LoRA:
r=16
lora_alpha=16
lora_dropout=0.05
bias="none"
task_type="CAUSAL_LM"
target_modules=['k_proj', 'gate_proj', 'v_proj', 'up_proj', 'q_proj', 'o_proj', 'down_proj']
Training arguments:
per_device_train_batch_size=4
gradient_accumulation_steps=4
gradient_checkpointing=True
learning_rate=5e-5
lr_scheduler_type="cosine"
max_steps=200
optim="paged_adamw_32bit"
warmup_steps=100
DPOTrainer:
beta=0.1
max_prompt_length=1024
max_length=1536
NeuralHermes is based on the model that has been further fine-tuned with Direct Preference Optimization (DPO) using the dataset. It surpasses the original model on most benchmarks (see results).
It is directly inspired by the RLHF process described by 's authors to improve performance. I used the same dataset and reformatted it to apply the ChatML template.
The code to train this model is available on and . It required an A100 GPU for about an hour.
GGUF:
AWQ:
GPTQ:
3.0bpw:
4.0bpw:
5.0bpw:
6.0bpw:
8.0bpw:
Teknium (author of OpenHermes-2.5-Mistral-7B) benchmarked the model ().
You can check the Weights & Biases project .
You can use this easy to use and cheap LLM Api here at