Multi-tool agent workflows with OmniCoder
#12
by O96a - opened
Running this on 4-bit quant with Ollama for code generation tasks — the image-text-to-text capability is interesting for multi-modal agent contexts. Curious if anyone has benchmarked the code completion accuracy against dedicated code models like Codestral or Qwen2.5-Coder at similar sizes? The agent tag suggests it's tuned for tool-calling workflows, but the model card doesn't specify the tool format. Is it native function calling or structured output?