liamka/sf-100

TEXT GENERATIONConcurrent Unit Cost:1Model Size:7.6BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Apr 2, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

liamka/sf-100 is a 7.6 billion parameter conversational language model developed by liamka, fine-tuned from the Qwen2.5-7B-Instruct architecture. This model is specifically supervised fine-tuned for general-purpose single- and multi-turn chat applications. It leverages a BF16 merged weight precision, built upon a 4-bit BNB-quantized base, and is designed for conversational assistant tasks.

Loading preview...

Model Overview

liamka/sf-100 is a 7.6 billion parameter conversational language model, developed by liamka, based on the Qwen2.5-7B-Instruct architecture. It has been supervised fine-tuned (SFT) using Hugging Face TRL, specifically for chat-based interactions. The model utilizes BF16 merged weights, trained on top of a 4-bit bnb-quantized base, and is released under the Apache-2.0 license.

Key Capabilities

  • General-purpose conversational assistant: Designed for both single-turn and multi-turn chat scenarios.
  • Multi-language support: While not formally evaluated for non-English input, the base Qwen2.5 model supports multiple languages.
  • Efficient architecture: Built on a 4-bit quantized base for potentially more efficient deployment.

Training Details

The model was trained using Unsloth and TRL's SFTTrainer for supervised fine-tuning. Notably, it does not incorporate RLHF or DPO, meaning its safety and refusal behaviors are primarily inherited from its base model or are weaker.

Limitations and Considerations

  • Inherits biases and knowledge cutoff from the Qwen2.5-7B-Instruct base model.
  • No preference optimization (RLHF/DPO): Safety and refusal behaviors may not be as robust as models with preference alignment.
  • Potential for hallucination: Factual claims generated by the model should always be verified.
  • Informal evaluation: No formal benchmark numbers are reported, and non-English input has not been evaluated.

Intended Use Cases

  • Single- and multi-turn chat applications.

Not Suitable For

  • Safety-critical settings (e.g., medical, legal, financial advice).
  • High-stakes factual lookup without external verification.

Finetunes
1 models