seihyun87/broken-model

TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:32kPublished:Mar 29, 2026Architecture:Transformer Cold

seihyun87/broken-model is a corrected version of yunmorning/broken-model, based on the Qwen3-8B architecture. This model addresses a critical issue where the /chat/completions API endpoint failed due to a missing chat_template in its tokenizer configuration. By integrating the official Qwen3 chat template, it now properly supports chat-based interactions, including system, user, assistant, tool call, and reasoning block formatting. This fix enables the model to be used effectively for conversational AI applications.

Loading preview...

broken-model: Corrected Qwen3-8B for Chat API

This model, seihyun87/broken-model, is a critical fix for the original yunmorning/broken-model, which was based on the Qwen3-8B architecture. The primary issue prevented the /chat/completions API endpoint from functioning, returning an error indicating "no configured chat prompt template."

Key Capabilities

  • Enabled Chat API: The core fix involves adding the official Qwen/Qwen3-8B chat_template to tokenizer_config.json.
  • Standard ChatML Support: Now correctly processes messages arrays for chat interactions.
  • Advanced Formatting: Supports system, user, and assistant roles, along with specific formatting for tool calls (<tool_call>/</tool_call>) and reasoning/thinking blocks (<think>/</think>).
  • Accurate Metadata: The README.md has been updated to correctly reflect Qwen/Qwen3-8B as the base model, preventing user confusion.

Good for

  • Developers requiring a functional /chat/completions endpoint for Qwen3-8B based models.
  • Building conversational AI applications that leverage Qwen3's capabilities.
  • Use cases involving tool integration or explicit reasoning steps within chat interactions.

Finetunes
1 models