osola / Affine-new_v23_1

Affine-eim_hades Improved Model

This is an improved version of the ben0925/Affine-eim_hades model, fine-tuned on combined Affine datasets for enhanced performance.

Model Details

• Base Model: Affine-eim_hades (7B parameters, Mistral-based)
• Model Type: Causal Language Model
• Training Approach: Full fine-tuning
• Hardware: 4x NVIDIA H200 GPUs
• Precision: BFloat16
• Framework: DeepSpeed ZeRO-3

Training Data

Combined dataset from multiple Affine sources:

• Affine-Ded_ABD_2500_slight (23.3k examples)
• Affine-Ded_Abd_slight_Trongg (23.3k examples)
• Affine-Ded_1500 (23.3k examples)
• Additional smaller datasets

Total training examples: ~70k+

Training Configuration

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

# Load model and tokenizer
model = AutoModelForCausalLM.from_pretrained(
    "path/to/model",
    torch_dtype=torch.bfloat16,
    device_map="auto",
    trust_remote_code=True
)

tokenizer = AutoTokenizer.from_pretrained("path/to/model")

# Generate
prompt = "Your prompt here"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

with torch.no_grad():
    outputs = model.generate(
        **inputs,
        max_new_tokens=512,
        temperature=0.7,
        top_p=0.9,
        do_sample=True
    )

response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)

Improvements Over Base Model

This improved version was trained using:

• Multi-GPU distributed training on H200s
• Advanced optimization techniques (DeepSpeed, gradient checkpointing)
• Combined high-quality Affine datasets
• Optimized hyperparameters for better convergence

Performance

See evaluation results in the results/ directory for detailed benchmarks comparing this model to the original Affine-eim_hades.

Citation

@misc{{affine-eim-hades-improved,
  title={{Affine-eim_hades Improved Model}},
  author={{Your Name}},
  year={{2025}},
  publisher={{HuggingFace}},
  howpublished={{\url{{https://huggingface.co/your-model-path}}}}
}}

License

Apache 2.0 (following base model license)

Acknowledgments

• Original model: ben0925/Affine-eim_hades
• Datasets: ben0925's Affine dataset collection
• Framework: Affine Foundation