ClickForge 2.7 — fix local thinking-model responses
b668aa3
This commit was signed with the committer’s verified signature.
stevologic
Stephen M Abbott
Bug fix for the local (Ollama) provider.
Fixed: using a local "thinking" model (e.g. qwen3.5:4b) returned "the model returned an empty response." Those models were spending their whole token budget on internal reasoning before emitting the answer. ClickForge now tells Ollama not to think (think: false) so the model answers directly — confirmed responding in ~2s instead of failing. Small/quantized models that emit slightly malformed JSON are also now tolerated.
Tip: tiny non-thinking models like qwen2.5:0.5b are the fastest for this; you can grab one with Download model in the AI tab.
SHA-256 of ClickForge.exe:
820832D8ADA8B995F18B194909953A43DCA026C318A7B8E3C4FF166146A76E8D
Verify with: Get-FileHash .\ClickForge.exe -Algorithm SHA256
Free and open source under the MIT License.