ramgpt/qwen3.5-9b-factorio-build-sft-GGUF overview
qwen3.5 9b factorio build sft GGUF GGUF conversion of divagr1925/qwen3.5 9b factorio build sft . Files qwen3.5 9b factorio build sft Q4 K M.gguf — Q4 K M — 5.2…
Runs locally from ~5.24 GB disk (8 GB VRAM class GPUs with llama.cpp / guIDE).
Repository Files & Downloads
| File | Type | Quantization | Size | Link |
|---|---|---|---|---|
| qwen3.5-9b-factorio-build-sft-Q4_K_M.gguf | GGUF | Q4_K_M | 5.24 GB | Download |
Model Details
Model README
---
base_model: divagr1925/qwen3.5-9b-factorio-build-sft
tags:
- gguf
- llama.cpp
---
qwen3.5-9b-factorio-build-sft-GGUF
GGUF conversion of divagr1925/qwen3.5-9b-factorio-build-sft.
Files
qwen3.5-9b-factorio-build-sft-Q4_K_M.gguf— Q4_K_M — 5.24 GiB
SHA256: f58eec68efd9b64a3063220615c0de8f0a44f306f6f2017885b91167378b6f82
Conversion scope
The source checkpoint is multimodal, but this GGUF is a text-only conversion. Vision/image weights and an mmproj artifact are not included. Do not treat this file as preserving the source model's multimodal capability.
Conversion note
Conversion used --no-mtp because source config and source tensors disagreed about MTP/NextN presence.
Validation gate
Before upload this exact file passed a real llama-server /v1/chat/completions smoke gate using the source chat template path. The gate checks:
- model loads successfully
- system/user chat template separation
- exact-output compliance for strict smoke prompts
- generation ends with finish_reason=stop instead of hitting the token cap
- no pseudo-role continuation such as /user or /assistant
- basic arithmetic generation
- no raw chat control-token leakage
- no thinking-tag leakage when reasoning is disabled
Reasoning mode used for smoke: off.
Task-specific verifier
Verifier: factory-sim 0.1.2 factorio_build.core.score_completion.
Observed Q4 result: 0/128 scene-runs succeeded (0.00%). Sandbox-valid programs: 11/16. Programs with non-zero success: 0/16. Programs hitting the completion cap: 1/16.
Settings: split=val, rows=4, samples/row=4, scenes/row=8, temperature=0.6, max_completion_tokens=12288, thinking=False.
Caveat: Source model card reports 18.0% validation using 16 rows x 4 samples at temperature 0.6. This Q4 check used only 4 rows x 4 samples, so the rates are not directly comparable; the observed Q4 result was 0/128 scene-runs.
Source
Source revision used for conversion: 71c26b72a4df85ca909b3e97d9c231356a74c039.
Source HEAD checked immediately before publish: bb3192edcd3d05e712db6e486da0077699ce0927.
If these revisions differ while the source safetensors fingerprints are unchanged, the difference is metadata-only and publishing remains allowed.
All model credit belongs to the original model authors.
Run ramgpt/qwen3.5-9b-factorio-build-sft-GGUF with guIDE
Download guIDE — the AI-native code editor with local LLM inference and 69 built-in tools.
Source: Hugging Face · Compare models