granite-8b-code-instruct-4k-FP8

granite-8b-code-instruct-4k-FP8

FP8 quantized version of IBM's Granite 8B Code model for efficient inference

This is an FP8 (E4M3) quantized version of ibm-granite/granite-8b-code-instruct-4k using compressed_tensors format. Quantized by TevunahAi on enterprise-grade hardware.

🎯 Recommended Usage: vLLM

For optimal performance with full FP8 benefits (2x memory savings + faster inference), use vLLM or TensorRT-LLM:

Quick Start with vLLM

pip install vllm

Python API:

from vllm import LLM, SamplingParams

# vLLM auto-detects FP8 from model config
llm = LLM(model="TevunahAi/granite-8b-code-instruct-4k-FP8", dtype="auto")

# Generate
prompt = "Write a Python function to calculate fibonacci numbers:"
sampling_params = SamplingParams(temperature=0.7, max_tokens=256)

outputs = llm.generate([prompt], sampling_params)
for output in outputs:
    print(output.outputs[0].text)

OpenAI-Compatible API Server:

vllm serve TevunahAi/granite-8b-code-instruct-4k-FP8 \
    --dtype auto \
    --max-model-len 4096

Then use with OpenAI client:

from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:8000/v1",
    api_key="token-abc123",  # dummy key
)

response = client.chat.completions.create(
    model="TevunahAi/granite-8b-code-instruct-4k-FP8",
    messages=[
        {"role": "user", "content": "Write a Python function to calculate fibonacci numbers"}
    ],
    temperature=0.7,
    max_tokens=256,
)

print(response.choices[0].message.content)

vLLM Benefits

• ✅ Weights, activations, and KV cache in FP8
• ✅ ~8GB VRAM (50% reduction vs BF16)
• ✅ Native FP8 tensor core acceleration on Ada/Hopper GPUs
• ✅ Faster inference with optimized CUDA kernels
• ✅ Runs on consumer GPUs (RTX 4070, RTX 4060 Ti 16GB, RTX 5000 Ada)

⚙️ Alternative: Transformers

This model can also be loaded with transformers. Note: Transformers will decompress FP8 → BF16 during inference. However, at 8B parameters, this is manageable (~16GB VRAM).

<details> <summary>Transformers Example (Click to expand)</summary>

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

# Loads FP8 weights but decompresses to BF16 during compute
model = AutoModelForCausalLM.from_pretrained(
    "TevunahAi/granite-8b-code-instruct-4k-FP8",
    device_map="auto",
    torch_dtype="auto",
    low_cpu_mem_usage=True,
)
tokenizer = AutoTokenizer.from_pretrained("TevunahAi/granite-8b-code-instruct-4k-FP8")

# Generate
prompt = "Write a Python function to calculate fibonacci numbers:"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Requirements:

pip install torch>=2.1.0 transformers>=4.40.0 accelerate compressed-tensors

System Requirements:

• ~16GB VRAM (decompressed to BF16)
• CUDA 11.8 or newer
• PyTorch 2.1+ with CUDA support

</details>

📊 Quantization Details

Quantization Infrastructure

Professional hardware ensures consistent, high-quality quantization:

• CPUs: Dual Intel Xeon Max 9480 (112 cores / 224 threads, 128GB HBM2e)
• GPU: NVIDIA RTX 5000 Ada Generation (32GB VRAM, native FP8 support)
• Memory: 256GB DDR5 + 128GB HBM2e = 384GB total system memory
• Software Stack: Ubuntu 25.10 | Python 3.12 | PyTorch 2.8 | CUDA 13.0 | llm-compressor

🔧 Why FP8?

With vLLM/TensorRT-LLM:

• ✅ 50% memory reduction vs BF16 (weights + activations + KV cache)
• ✅ Faster inference via native FP8 tensor cores
• ✅ Better throughput with optimized kernels
• ✅ Minimal quality loss for code generation tasks
• ✅ Accessible on consumer GPUs (RTX 4060 Ti 16GB+)

With Transformers:

• ✅ Smaller download size (~8GB vs ~16GB BF16)
• ✅ Compatible with standard transformers workflow
• ⚠️ Decompresses to BF16 during inference (no runtime memory benefit)

For production inference, use vLLM to realize the full FP8 benefits.

💾 Model Files

This model is sharded into multiple safetensors files (all required for inference). The compressed format enables efficient storage and faster downloads.

🔬 IBM Granite Code Models

Granite Code models are specifically trained for code generation, editing, and explanation tasks. This 8B parameter version offers strong performance on:

• Code completion and generation
• Bug fixing and refactoring
• Code explanation and documentation
• Multiple programming languages
• 4K context window

Granite 8B vs Larger Models:

• ✅ Fast iteration - quick response times
• ✅ Accessible - runs on consumer GPUs
• ✅ Good quality - suitable for most coding tasks
• ⚠️ Trade-off: Less capable on very complex reasoning vs 20B/34B

📚 Original Model

This quantization is based on ibm-granite/granite-8b-code-instruct-4k by IBM.

For comprehensive information about:

• Model architecture and training methodology
• Supported programming languages
• Evaluation benchmarks and results
• Ethical considerations and responsible AI guidelines

Please refer to the original model card.

🔧 Hardware Requirements

Minimum (vLLM):

• GPU: NVIDIA RTX 4060 Ti (16GB) or better
• VRAM: 8GB minimum, 12GB+ recommended
• CUDA: 11.8 or newer

Recommended (vLLM):

• GPU: NVIDIA RTX 4070 / 4090 / RTX 5000 Ada
• VRAM: 12GB+
• CUDA: 12.0+

Transformers:

• GPU: Any CUDA-capable GPU
• VRAM: 16GB+
• Works but not optimal for performance

📖 Additional Resources

• vLLM Documentation: docs.vllm.ai
• TensorRT-LLM: github.com/NVIDIA/TensorRT-LLM
• TevunahAi Models: huggingface.co/TevunahAi
• llm-compressor: github.com/vllm-project/llm-compressor
• IBM Granite: github.com/ibm-granite

📄 License

This model inherits the Apache 2.0 License from the original Granite model.

🙏 Acknowledgments

• Original Model: IBM Granite team
• Quantization Framework: Neural Magic's llm-compressor
• Quantized by: TevunahAi

📝 Citation

If you use this model, please cite the original Granite work:

@misc{granite2024,
  title={Granite Code Models},
  author={IBM Research},
  year={2024},
  url={https://huggingface.co/ibm-granite/granite-8b-code-instruct-4k}
}

---

<div align="center">

Professional AI Model Quantization by TevunahAi

Enterprise-grade quantization on specialized hardware

View all models | Contact for custom quantization

</div>