JabirTech/choqok-1B-0.0-alpha-1

Hugging Face
TEXT GENERATIONConcurrent Unit Cost:1Model Size:1BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jan 8, 2025License:mitArchitecture:Transformer0.0K Open Weights Featherless Exclusive Warm

JabirTech/choqok-1B-0.0-alpha-1 is a small language model (SLM) based on Meta's LLaMA 3.2 1B architecture, fine-tuned specifically on Persian data. Developed by Muhammadreza Haghiri as part of the Jabir Project, this model is designed for on-device applications and excels at generating Persian content. It is freely and commercially usable, aiming to provide democratized AI products with a focus on Persian language support.

Loading preview...

Choqok 1B (Version 0.0-alpha-1) Overview

Choqok is a small language model (SLM) developed by Muhammadreza Haghiri under the Jabir Project, aiming to democratize AI products. This initial alpha version is based on Meta's LLaMA 3.2 1B architecture and has been specifically fine-tuned on Persian data.

Key Capabilities & Features

  • Persian Language Focus: Optimized for generating content in Persian, making it suitable for applications requiring strong Persian language understanding and generation.
  • On-Device Design: Named "Choqok" (Persian for a small bird) to reflect its compact size, it is designed for efficient on-device deployment, requiring approximately 3GB of VRAM.
  • Open Usage: The model is released under a permissive license, allowing for free and commercial use.
  • Integration: Provides clear instructions for integration using the transformers library in Python and offers an OpenAI-compatible API and a dedicated Chat UI for interaction.

Known Issues & Development Focus

As an alpha release, the developers are actively working on improving several aspects:

  • Persian Coherence: Enhancing the quality and coherence of Persian responses.
  • Tokenization: Addressing issues with tokenization that may lead to the presence of non-Persian characters (e.g., Chinese, Russian, Thai, Hebrew) in Persian or Arabic content.
  • Model Identity: Acknowledging that the model may still identify itself as LLaMA due to its fine-tuning origin.

Future updates will include the release of the dataset used for training and further improvements to the model's performance and tokenizer.

Finetunes
1 models