MetalGPT-1-GGUF

MetalGPT-1 GGUF

This repository contains unofficial GGUF conversions of the nn-tech/MetalGPT-1 model for use with GGUF-compatible runtimes.

MetalGPT-1 is a 32B chat model based on Qwen/Qwen3-32B, further trained with both continual pre-training and supervised fine-tuning on domain-specific data from the mining and metallurgy industry.

⚠️ Disclaimer:
This repository is not affiliated with the original authors of MetalGPT-1.
These are pure quantizations of the original model weights - no additional training, fine-tuning, or modifications were applied.
Quality, correctness, and safety of the quantized variants are not guaranteed.

See the original model card: https://huggingface.co/nn-tech/MetalGPT-1

---

GGUF variants in this repository

The following GGUF quantized variants of MetalGPT-1 are provided:

Choose a variant based on your hardware and quality requirements:

• Q4_K_M / Q4_K_S: best options for low‑VRAM environments.
• Q6_K / Q8_0: better fidelity for demanding generation quality.

Note: Try adding the /think tag to your prompts if you want to explicitly trigger reasoning capabilities.

VRAM guidance

These numbers are rough rules of thumb for 32B GGUF inference; actual VRAM/RAM usage depends on runtime/backend, context size (KV cache), and overhead.

• < 24 GB VRAM: you’ll likely need partial GPU offload (some weights/layers stay in system RAM). Prefer Q4_K_M / Q4_K_S.
• ~24 GB VRAM: Q4 variants typically fit best; higher quants may still require partial offload depending on context size.
• ~32 GB VRAM: Q6_K is a reasonable target; may still require tuning/offload for large contexts.
• 40 GB+ VRAM: Q8_0 is usually the go-to “max fidelity quant” option among the listed files.
• 80 GB+ VRAM: consider running the original (non-quantized) weights instead of quants if you want maximum fidelity.

Note: partial offload (keeping some layers in system RAM) can significantly reduce throughput vs full GPU offload.

---

Usage with LM Studio

1. Download LM Studio from here.
2. Search for "NuisanceValue/MetalGPT-1-GGUF" in the model hub within LM Studio.
3. Select a quantization variant.
4. Once downloaded, select the model in the menu.

Usage with Ollama

1. Install Ollama from the official website and ensure the ollama command is available in your terminal.
2. In the terminal, run the model directly from Hugging Face (you can specify the desired quantization tag after a colon):

   ollama run hf.co/NuisanceValue/MetalGPT-1-GGUF:Q4_K_M
   

3. After the first run, the model will appear in your local model list:

   ollama list
   

Note: You can also use Ollama through a web UI such as OpenWebUI by configuring it to connect to your Ollama server.

Usage with llama.cpp

Download one of the GGUF files (for example MetalGPT-1-32B-Q4_K_M.gguf) and run:

./llama-cli \
  -m MetalGPT-1-32B-Q4_K_M.gguf \
  -p "Назови плюсы и минусы хлоридной и сульфатной технологии производства никеля." \
  --temp 0.7 \
  --top-p 0.8 \
  --top-k 70 \
  --n-predict 512 \
  --ctx-size 8192

Tip (GPU offload): you can add -ngl N (aka --n-gpu-layers) — it controls how many layers are offloaded to VRAM, while the rest stays in system RAM. Start with -ngl -1 (try to offload all layers); if you hit an out-of-memory error, lower it (e.g., -ngl 20, -ngl 30, …) until it fits.

Usage with llama-cpp-python

Install llama-cpp-python if you haven't already:

pip install llama-cpp-python

Then use the following code snippet to load the model and generate text:

from llama_cpp import Llama

# Path to your GGUF file
model_path = "MetalGPT-1-32B-Q4_K_M.gguf"

# Initialize the model
llm = Llama(
    model_path=model_path,
    n_gpu_layers=-1,      # Offload all layers to GPU. If you get an OOM error, change this number to offload some layers to RAM (e.g., to 20 or 30).
    n_ctx=8192,           # Context window (adjust based on VRAM)
    verbose=False
)

messages = [
    {"role": "system", "content": "Ты специалист в области металлургии."},
    {"role": "user", "content": "Назови плюсы и минусы хлоридной и сульфатной технологии производства никеля."},
]

output = llm.create_chat_completion(
    messages=messages,
    max_tokens=2048,
    temperature=0.7,
    top_p=0.8
)

print(output["choices"][0]["message"]["content"])