---

license: apache-2.0

base_model: org2ai/Wald-4B

base_model_relation: quantized

pipeline_tag: text-generation

language:

• en

tags:

• gguf
• llama.cpp
• decision-model
• calibrated-probabilities
• wald
• wald-q4b
• qwen3.5
• 4b

---

Wald-4B GGUF

GGUF files of Wald-Q4B v2.1 (tag v2.1) for llama.cpp: an open 4B decision model that returns a calibrated probability for every option.

| File | Size | Same answer as BF16 (JevBench public, one pass) |

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

| Wald-4B-v2.1-Q8_0.gguf | 4.5 GB | 231/231 |

| Wald-4B-v2.1-Q6_K.gguf | 3.5 GB | 226/231 |

| Wald-4B-v2.1-Q5_K_M.gguf | 3.1 GB | 226/231 |

| Wald-4B-v2.1-Q4_K_M.gguf | 2.7 GB | 225/231 |

| Wald-4B-v2-mmproj-F16.gguf (vision projector for llama.cpp's multimodal tools; the vision tower is unchanged since v2, so this file serves v2.1 too) | 0.67 GB | |

Quick start

hf download org2ai/Wald-4B-GGUF --include "Wald-4B-v2.1-Q4_K_M.gguf" "v2.1/*" --local-dir ./wald-gguf
pip install "./wald-gguf/v2.1/server" transformers
wald-serve-native --gguf ./wald-gguf/Wald-4B-v2.1-Q4_K_M.gguf --tokenizer-dir ./wald-gguf/v2.1 --effort none --port 8000

How it works

• One pass, text only: the bundled server reads a probability for every option from the model's option-letter logits (Qwen chat template), calibrated with the same frozen temperature table as the BF16 release.
• --effort auto (Auto 0.7) also runs here but has not been parity-checked on GGUF; for images, use the BF16 release.
• On a 1,000-request Decision Index sample (5,922 questions), Q8_0 gives the BF16 answer on 99.5 % of questions and Q4_K_M on 97.3 %. All numbers: evaluation/v2.1-gguf-parity.json.

Earlier versions

• v2 (Wald-4B-v2-*.gguf, v2/): Wald-Q4B v2, unchanged in this repository.
• v1.2 (Wald-4B-v1.2-*.gguf, server/): unchanged in this repository.

Apache-2.0.