{ "validation_type": "GGUF binary header and structural-count validation", "runtime_note": "A CPU-only llama.cpp generation test on TRUBA exceeded a five-minute startup timeout for Q8_0. Quant-specific runtime and JSON benchmarks are therefore not claimed.", "results": [ { "quantization": "Q8_0", "file": "Qwen3-1.7B-ResearchReasoning-JSON-RL-Q8_0.gguf", "size_bytes": 1834426080, "magic": "GGUF", "gguf_version": 3, "tensor_count": 310, "metadata_kv_count": 29, "structural_header_ok": true }, { "quantization": "Q5_K_M", "file": "Qwen3-1.7B-ResearchReasoning-JSON-RL-Q5_K_M.gguf", "size_bytes": 1257879264, "magic": "GGUF", "gguf_version": 3, "tensor_count": 310, "metadata_kv_count": 29, "structural_header_ok": true }, { "quantization": "Q4_K_M", "file": "Qwen3-1.7B-ResearchReasoning-JSON-RL-Q4_K_M.gguf", "size_bytes": 1107408608, "magic": "GGUF", "gguf_version": 3, "tensor_count": 310, "metadata_kv_count": 29, "structural_header_ok": true } ] }