seanpoyner/smolcode-coder-terraform-1.5b-tools
The seanpoyner/smolcode-coder-terraform-1.5b-tools is a 1.5 billion parameter LoRA fine-tune of Qwen2.5-Coder-1.5B-Instruct, developed by seanpoyner. This model is specifically trained to emit native function calls, enabling agentic coding loops for small language models. With a context length of 32768 tokens, it excels at integrating tool use within coding assistants, particularly for Terraform-related tasks.
Loading preview...
Overview
This model, seanpoyner/smolcode-coder-terraform-1.5b-tools, is a 1.5 billion parameter LoRA fine-tune of the Qwen/Qwen2.5-Coder-1.5B-Instruct base model. Its primary purpose is to enable small language models (SLMs) to drive agentic coding loops by correctly emitting native <tool_call> function calls, which standard Qwen-Coder models typically describe as plain-text JSON.
Key Capabilities & Training
- Native Tool Calling: Fine-tuned to produce
<tool_call>format for agentic tool use, crucial for integration with runtimes like Ollama and llama.cpp. - Base Model: Utilizes
Qwen/Qwen2.5-Coder-1.5B-Instructas its foundation. - Training Method: Employed bf16 LoRA (r=16, α=32) on attention and MLP projections, with assistant-only loss focused on tool calls and final answers.
- Training Data: Leveraged
NousResearch/hermes-function-calling-v1for breadth and syntheticsmolcodetool-use trajectories for specific sharpness, all rendered with theapply_chat_template(tools=...)for byte-identical training targets. - Context Length: Supports a 32768 token context length.
Use Cases
This model is designed for use with the standard Qwen2.5 chat template, responding with <tool_call>{"name": ..., "arguments": ...}</tool_call> when tool interaction is required. It is particularly suited for:
- Agentic coding assistants that require precise tool invocation.
- Applications where small, efficient coder models need to interact with external tools.
- Environments where native tool call parsing is essential for workflow automation.