hamishivi/tmax-qwen36-27b-tmax-sft-full-20260513-epoch1

VISIONConcurrency Cost:2Model Size:27BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:May 19, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

The hamishivi/tmax-qwen36-27b-tmax-sft-full-20260513-epoch1 is a 27 billion parameter causal language model, derived from hamishivi/Qwen3.6-27B and fine-tuned using the open-instruct framework. This intermediate SFT checkpoint represents the model state after one full epoch of training. It is designed for general-purpose language generation and understanding tasks, leveraging its Qwen3.6 base architecture.

Loading preview...

Model Overview

The hamishivi/tmax-qwen36-27b-tmax-sft-full-20260513-epoch1 is a 27 billion parameter large language model. It is an intermediate supervised fine-tuning (SFT) checkpoint based on the hamishivi/Qwen3.6-27B model, processed through the open-instruct framework.

Key Characteristics

  • Base Model: Derived from hamishivi/Qwen3.6-27B.
  • Training State: This specific release is a snapshot of the model after completing epoch 1 of its fine-tuning process.
  • Conversion Process: The model was converted from a DeepSpeed ZeRO-2 sharded BF16 checkpoint to consolidated FP32, then cast back to BF16, and finally saved in the Hugging Face safetensors format.
  • Tokenizer: Utilizes the tokenizer and chat template from the original hamishivi/Qwen3.6-27B model.

Usage

This model can be loaded and used like any standard Hugging Face causal language model, supporting common inference workflows with AutoModelForCausalLM and AutoTokenizer.