---

license: mit

base_model: ProCreations/grug-35b

base_model_relation: quantized

library_name: gguf

pipeline_tag: text-generation

tags:

• gguf
• code
• agent
• tool-use
• reasoning
• grug

---

grug-35b GGUF

grug brain inside rock. no custom Grug instruction required.

> July 15, 2026 GGUF replacement: same public repo, all popular quants rebuilt from

> the corrected parent. Old files preserved on pre-intrinsic-style-fix-2026-07-14.

These files were rebuilt from the intrinsic-style-gated

ProCreations/grug-35b checkpoint. The parent and

embedded template contain no hidden Grug/style prompt; ordinary chat and tool-enabled

agent scaffolds rely on the tuned weights.

27b and 35b hunt same prey

both parent grug hunt HumanEval and sanitized MBPP. number below come from big

parent brain, NOT squeezed GGUF rock. grug not claim rock test it never get.

number show pass@1 percent. bold grug win that hunt.

| hunt | grug-27b v2.1 | grug-35b rebuilt |

|---|---:|---:|

| HumanEval (164) | 87.2 | 80.5 |

| MBPP sanitized (100) | 85.0 | 88.0 |

| file | size |

|---|---:|

| grug-35b-Q4_K_M.gguf | 19.71 GiB |

| grug-35b-Q5_K_M.gguf | 23.03 GiB |

| grug-35b-Q6_K.gguf | 26.56 GiB |

| grug-35b-Q8_0.gguf | 34.37 GiB |

• Q4_K_M: recommended local default
• Q5_K_M: more accuracy meat
• Q6_K: high-quality rock
• Q8_0: closest popular quantized weight

All four loaded and generated successfully with llama.cpp f955e394bf94e01e5e36186d13c985727e5ef5b5 before upload.

Hashes live in SHA256SUMS; smoke evidence lives in smoke-results.json.

Full BF16 parent: HumanEval 80.5%, MBPP

88.0%, broad valid/strict/right

100.0/100.0/95.0%,

unprompted dialect-clean 100.0%.

Quant-specific full benchmark not claimed. Runtime must honor the embedded Qwen3.5

template. Reasoning stays in <think>...</think>; native XML tool calls stay intact.