---
license: creativeml-openrail-m
library_name: transformers
tags:
- deep_think
- reasoning
- chain_of_thought
- chain_of_thinking
- prev_2
- self_reasoning
- llama-cpp
- gguf-my-repo
language:
- en
base_model: prithivMLmods/Llama-Thinker-3B-Preview2
pipeline_tag: text-generation
---

# Triangle104/Llama-Thinker-3B-Preview2-Q4_K_S-GGUF
This model was converted to GGUF format from [`prithivMLmods/Llama-Thinker-3B-Preview2`](https://huggingface.co/prithivMLmods/Llama-Thinker-3B-Preview2) using llama.cpp via the ggml.ai's [GGUF-my-repo](https://huggingface.co/spaces/ggml-org/gguf-my-repo) space.
Refer to the [original model card](https://huggingface.co/prithivMLmods/Llama-Thinker-3B-Preview2) for more details on the model.

---
Model details:
-
Llama-Thinker-3B-Preview2 is a pretrained and instruction-tuned 
generative model designed for multilingual applications. These models 
are trained using synthetic datasets based on long chains of thought, 
enabling them to perform complex reasoning tasks effectively.


Model Architecture: [ Based on Llama 3.2 ] is an autoregressive 
language model that uses an optimized transformer architecture. The 
tuned versions undergo supervised fine-tuning (SFT) and reinforcement 
learning with human feedback (RLHF) to align with human preferences for 
helpfulness and safety.



	
		
	

		Use with transformers
	



Starting with transformers >= 4.43.0 onward, you can run conversational inference using the Transformers pipeline abstraction or by leveraging the Auto classes with the generate() function.


Make sure to update your transformers installation via pip install --upgrade transformers.


import torch
from transformers import pipeline

model_id = "prithivMLmods/Llama-Thinker-3B-Preview2"
pipe = pipeline(
    "text-generation",
    model=model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto",
)
messages = [
    {"role": "system", "content": "You are a pirate chatbot who always responds in pirate speak!"},
    {"role": "user", "content": "Who are you?"},
]
outputs = pipe(
    messages,
    max_new_tokens=256,
)
print(outputs[0]["generated_text"][-1])



Note: You can also find detailed recipes on how to use the model locally, with torch.compile(), assisted generations, quantised and more at huggingface-llama-recipes



	
		
	

		Use with llama
	



Please, follow the instructions in the repository


To download Original checkpoints, see the example command below leveraging huggingface-cli:


huggingface-cli download prithivMLmods/Llama-Thinker-3B-Preview2 --include "original/*" --local-dir Llama-Thinker-3B-Preview2



Here’s a version tailored for the Llama-Thinker-3B-Preview2-GGUF model:




	
		
	

		How to Run Llama-Thinker-3B-Preview2 on Ollama Locally
	



This guide demonstrates how to run the Llama-Thinker-3B-Preview2-GGUF
 model locally using Ollama. The model is instruction-tuned for 
multilingual tasks and complex reasoning, making it highly versatile for
 a wide range of use cases. By the end, you'll be equipped to run this 
and other open-source models with ease.




	
		
	

		Example 1: How to Run the Llama-Thinker-3B-Preview2 Model
	



The Llama-Thinker-3B-Preview2 model is a pretrained 
and instruction-tuned LLM, designed for complex reasoning tasks across 
multiple languages. In this guide, we'll interact with it locally using 
Ollama, with support for quantized models.



	
		
	

		Step 1: Download the Model
	



First, download the Llama-Thinker-3B-Preview2-GGUF model using the following command:


ollama run llama-thinker-3b-preview2.gguf




	
		
	

		Step 2: Model Initialization and Download
	



Once the command is executed, Ollama will initialize and download the
 necessary model files. You should see output similar to this:


pulling manifest
pulling a12cd3456efg... 100% ▕██████████████████████████████████████████████████████████████████████████████████████████▏ 3.2 GB
pulling 9f87ghijklmn... 100% ▕██████████████████████████████████████████████████████████████████████████████████████████▏ 6.5 KB
verifying sha256 digest
writing manifest
removing any unused layers
success
>>> Send a message (/? for help)




	
		
	

		Step 3: Interact with the Model
	



Once the model is fully loaded, you can interact with it by sending prompts. For example, let's ask:


>>> How can you assist me today?



A sample response might look like this [may / maynot be identical]:


I am Llama-Thinker-3B-Preview2, an advanced AI language model designed to assist with complex reasoning, multilingual tasks, and general-purpose queries. Here are a few things I can help you with:

1. Answering complex questions in multiple languages.
2. Assisting with creative writing, content generation, and problem-solving.
3. Providing detailed summaries and explanations.
4. Translating text across different languages.
5. Generating ideas for personal or professional use.
6. Offering insights on technical topics.

Feel free to ask me anything you'd like assistance with!




	
		
	

		Step 4: Exit the Program
	



To exit the program, simply type:


/exit





	
		
	

		Example 2: Using Multi-Modal Models (Future Use)
	



In the future, Ollama may support multi-modal models where you can 
input both text and images for advanced interactions. This section will 
be updated as new capabilities become available.




	
		
	

		Notes on Using Quantized Models
	



Quantized models like llama-thinker-3b-preview2.gguf
 are optimized for efficient performance on local systems with limited 
resources. Here are some key points to ensure smooth operation:


VRAM/CPU Requirements: Ensure your system has adequate VRAM or CPU resources to handle model inference.
Model Format: Use the .gguf model format for compatibility with Ollama.




	
		
	

		Conclusion
	



Running the Llama-Thinker-3B-Preview2 model locally 
using Ollama provides a powerful way to leverage open-source LLMs for 
complex reasoning and multilingual tasks. By following this guide, you 
can explore other models and expand your use cases as new models become 
available.

---
## Use with llama.cpp
Install llama.cpp through brew (works on Mac and Linux)

```bash
brew install llama.cpp

```
Invoke the llama.cpp server or the CLI.

### CLI:
```bash
llama-cli --hf-repo Triangle104/Llama-Thinker-3B-Preview2-Q4_K_S-GGUF --hf-file llama-thinker-3b-preview2-q4_k_s.gguf -p "The meaning to life and the universe is"
```

### Server:
```bash
llama-server --hf-repo Triangle104/Llama-Thinker-3B-Preview2-Q4_K_S-GGUF --hf-file llama-thinker-3b-preview2-q4_k_s.gguf -c 2048
```

Note: You can also use this checkpoint directly through the [usage steps](https://github.com/ggerganov/llama.cpp?tab=readme-ov-file#usage) listed in the Llama.cpp repo as well.

Step 1: Clone llama.cpp from GitHub.
```
git clone https://github.com/ggerganov/llama.cpp
```

Step 2: Move into the llama.cpp folder and build it with `LLAMA_CURL=1` flag along with other hardware-specific flags (for ex: LLAMA_CUDA=1 for Nvidia GPUs on Linux).
```
cd llama.cpp && LLAMA_CURL=1 make
```

Step 3: Run inference through the main binary.
```
./llama-cli --hf-repo Triangle104/Llama-Thinker-3B-Preview2-Q4_K_S-GGUF --hf-file llama-thinker-3b-preview2-q4_k_s.gguf -p "The meaning to life and the universe is"
```
or 
```
./llama-server --hf-repo Triangle104/Llama-Thinker-3B-Preview2-Q4_K_S-GGUF --hf-file llama-thinker-3b-preview2-q4_k_s.gguf -c 2048
```