Hyperparameter	Value
LoRA Rank (r)	32
LoRA Alpha	64
LoRA Dropout	0.05
Learning Rate	2e-5
LR Scheduler	Cosine
Batch Size	4
Gradient Accumulation	2
Effective Batch Size	8
Max Sequence Length	1536
Epochs	1
Precision	bfloat16
Optimizer	AdamW

LoRA Target Modules

q_proj, k_proj, v_proj, o_proj
gate_proj, up_proj, down_proj

Training Results

Step	Eval Loss
500	0.5733
5,000	0.5803
10,000	0.5662
15,000	0.5651
21,000	0.5643

Usage

python

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("samarthraina/qwen2.5-7b-openhermes-v1")
tokenizer = AutoTokenizer.from_pretrained("samarthraina/qwen2.5-7b-openhermes-v1")

messages = [
    {"role": "user", "content": "Explain quantum computing in simple terms."}
]

text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt")

outputs = model.generate(**inputs, max_new_tokens=512, temperature=0.7, do_sample=True)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Hardware

GPU: NVIDIA RTX 4090 (48GB)
Training Time: ~21 hours

Limitations

Trained on 250k samples (subset of full OpenHermes-2.5)
Single epoch training
Conservative hyperparameters (lr=2e-5, r=32)
May not perform optimally on tasks outside the training distribution

License

This model is released under the Apache 2.0 license, following the base model's license.

Citation

If you use this model, please cite:

bibtex

@misc{qwen2.5-7b-openhermes-v1,
  author = {Samarth Raina},
  title = {Qwen2.5-7B OpenHermes V1},
  year = {2026},
  publisher = {HuggingFace},
  url = {https://huggingface.co/samarthraina/qwen2.5-7b-openhermes-v1}
}