FoodClockLabs-Etiquette-1-8B-GGUF
Etiquette 1 (8B) (
FoodClockLabs/Etiquette-1-8B) is an open-weight vision-language model released under Apache-2.0 by FoodClock Labs for offline food packaging and nutrition label extraction. Built by merging anr=32LoRA fine-tune intoQwen/Qwen3-VL-8B-Instruct, it was trained in BF16 on a single NVIDIA RTX PRO 6000 using 14,657 Open Food Facts products, 53,228 photos, and 60 packaging languages. It transcribes only visible, legible information without guessing, estimating, or converting values, producing structured JSON for product details, ingredients, nutrition tables, allergens, certifications, packaging, and photo tags. On the human-verified 100-productetiquette-printed-truthbenchmark, it achieves 83.0% overall agreement across 981 judgements, alongside 81.0% on a private 99-product holdout and 87.8% on a 100-product catalogue exam. Photo-tag recall reaches 100.0% for front images, 98.0% for nutrition images, and 89.9% for ingredient images. It scores 97.0% on kcal, 98.0% on salt, and 96.0% on sodium, with only 0–2 invented values, and achieves 94.0% on nutrition table basis, outperforming Claude Sonnet 5.5 (89.0%) and Claude Opus 5.5 (89.0%) on that metric. Overall results are competitive with Claude Sonnet 5.5 (87.4%) and Claude Opus 5.5 (90.8%), at roughly a tenth of a cent per pack in local electricity. Available as 17.5 GB merged BF16 weights for Transformers and GGUF quantizations forllama.cpp, Ollama, and LM Studio, it retains 81.1% overall accuracy inQ8_0and 79.9% inQ4_K_M, enabling local deployment on consumer GPUs and Macs.
Model Files
| File Name | Quant Type | File Size | File Link | Description |
|---|---|---|---|---|
| Etiquette-1-8B.BF16.gguf | BF16 | 16.4 GB | Link | Full BF16 weights. Highest quality, largest file size. |
| Etiquette-1-8B.Q3_K_L.gguf | Q3_K_L | 4.43 GB | Link | Lower quality but usable, good for low RAM availability. |
| Etiquette-1-8B.Q3_K_M.gguf | Q3_K_M | 4.12 GB | Link | Low quality. |
| Etiquette-1-8B.Q4_K_M.gguf | Q4_K_M | 5.03 GB | Link | Good quality, default size for most use cases, recommended. |
| Etiquette-1-8B.Q4_K_S.gguf | Q4_K_S | 4.8 GB | Link | Slightly lower quality with more space savings, recommended. |
| Etiquette-1-8B.Q5_K_M.gguf | Q5_K_M | 5.85 GB | Link | High quality, recommended. |
| Etiquette-1-8B.Q5_K_S.gguf | Q5_K_S | 5.72 GB | Link | High quality, recommended. |
| Etiquette-1-8B.Q6_K.gguf | Q6_K | 6.73 GB | Link | Very high quality, near perfect, recommended. |
| Etiquette-1-8B.Q8_0.gguf | Q8_0 | 8.71 GB | Link | Extremely high quality, generally unneeded but max available quant. |
| Etiquette-1-8B.mmproj-bf16.gguf | mmproj-bf16 | 1.16 GB | Link | Multimodal projection file in BF16 format. Used for vision/language models. |
llama.cpp
LLM inference in C/C++ — https://github.com/ggml-org/llama.cpp