robnav/fai_bm_fix2

TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:32kPublished:Mar 30, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

robnav/fai_bm_fix2 is an 8 billion parameter Qwen3-based language model, derived from Qwen/Qwen3-8B, with a 32768 token context length. This model provides a structural fix for metadata mismatches in yunmorning/broken-model, specifically by correcting the base model identification and integrating the necessary chat template. It is optimized to enable proper Chat Completions API functionality for the original broken model, ensuring correct prompt formatting for inference engines.

Loading preview...

Overview

robnav/fai_bm_fix2 is a specialized model repository designed to rectify critical metadata and configuration issues found in the yunmorning/broken-model. The primary goal of this fix is to enable the original model to function correctly with chat completion APIs, specifically addressing an error where the model lacked a configured chat prompt template.

Key Fixes and Analysis

  • Architecture Correction: The original model's README incorrectly cited meta-llama/Meta-Llama-3.1-8B, while the true base model, confirmed by config.json and tokenizer_config.json, is Qwen/Qwen3-8B.
  • Chat Template Integration: The core issue was the absence of a chat_template key in the local tokenizer_config.json, which is essential for the Chat Completions API to format messages. This repository introduces a chat_template.jinja file containing the official Qwen3 Jinja2 logic.
  • Metadata Alignment: The README.md YAML front matter has been updated to accurately reflect the Qwen3 base model.

Usage and Benefits

By using robnav/fai_bm_fix2, developers can now leverage the original yunmorning/broken-model with modern inference engines that automatically detect and apply the correct chat formatting. This ensures seamless integration with Chat Completions APIs, resolving the previous functionality limitations caused by the missing template.

Finetunes
1 models