appvoid/cloud-4
appvoid/cloud-4 is a 0.35 billion parameter language model created by appvoid using the DARE TIES merge method. It combines OrionLLM/Terminus-LFM2.5-350m and squ11z1/claude-oss-350m to integrate terminal/agentic capabilities with conversational instruction-following. This model is designed for applications requiring a blend of structured task execution and natural language interaction within a 32768-token context.
Loading preview...
Overview
appvoid/cloud-4 is a 0.35 billion parameter language model developed by appvoid, utilizing the DARE TIES (Disentangled Attribute Representation and Editing with TIES) merge method. This model was constructed by merging two distinct pre-trained language models, building upon MihaiPopa-1/LFM2.5-350M-heretic as its base.
Key Capabilities
This model is engineered to combine complementary strengths from its constituent models:
- Terminal/Agentic Behavior: Inherits capabilities from
OrionLLM/Terminus-LFM2.5-350m, making it suitable for tasks requiring structured execution, agentic reasoning, or tool interaction. - Conversational Instruction-Following: Integrates an assistant tone and proficiency in conversational instruction-following from
squ11z1/claude-oss-350m, enhancing its ability to engage in natural dialogue and respond to user prompts.
Merge Details
The DARE TIES method was applied with specific weighting and density parameters for each merged model. The configuration used bfloat16 for its dtype and included normalize: true and int8_mask: true in its parameters, aiming for an optimized blend of the source models' functionalities.
Good For
- Applications requiring a compact model that can handle both structured, agent-like tasks and provide natural, conversational responses.
- Scenarios where integrating terminal commands or tool use with user-friendly interaction is beneficial.
- Use cases that can leverage a 32768-token context length for processing longer inputs or maintaining conversational history.