LN-Tools — Evaluated open-weights families run in GGUF Q4_K_M: Qwen 2.5 (1.5b/3b/7b

Evaluated open-weights families run in GGUF Q4_K_M: Qwen 2.5 (1.5b/3b/7b; qwen2.5:3b is the canonical baseline), Llama 3.1:8B, Mistral 7B, Gemma 3 (1b/4b/12b), and DeepSeek-R1 (excluded for parser incompatibility, debt D17).

Source: PUMA project documentation · Traceability: corpus unit ANXAG-015 · Confidence: verified-at-primary-source

Related

• RES-MEM-025
• DEBT-PUMA-056