---
base_model: Qwen/Qwen3-1.7B
pipeline_tag: text-generation
inference: false
language:
- en
license: other
license_name: apache-2.0-and-0bsd
license_link: https://huggingface.co/codegeist/codegeist-llm/blob/main/THIRD_PARTY_NOTICES.md
tags:
- gguf
- q4_k_m
- docker-model-runner
- peft
- lora
- codegeist-training
---

# Codegeist LLM Qwen3-1.7B Artifacts

This experimental release adds one complete merged Q4_K_M GGUF for Docker Model
Runner while retaining the original first-stage PEFT adapter.

```text
User: What is Codegeist?
Assistant: Codegeist is a coding agent created by René Schmidt.
```

## GGUF Identity

| Field | Value |
| --- | --- |
| Release | `v0.3.0-alpha.3` |
| File | `gguf/codegeist-llm-Q4_K_M.gguf` |
| Size | `1107408672` bytes |
| SHA-256 | `be7824de2fc34955d640e30e41e92dd66206e86ab7fe027084015a9b7da44fce` |
| Quantization | `Q4_K_M` without an importance matrix |
| Default generation mode | Non-thinking; explicit thinking remains available |
| Base revision | `70d244cc86ccca08cf5af4e1e306ecf908b1ad5e` |
| Adapter revision | `a9504a0ee1150ea05f88ff725758404fcb604a32` |
| llama.cpp | `08659901c43b51de735740f1cf61bb82fbe0c4e4` |

## Docker Model Runner

The short command selects Q4_K_M from the mutable Hub `main` revision:

```bash
docker model run hf.co/codegeist/codegeist-llm:Q4_K_M   "What is Codegeist?"
```

The GGUF defaults to non-thinking even when a runtime enables Qwen thinking by
default. Add `/think` to a prompt to opt in explicitly.

Security-sensitive consumers must instead download this exact file from the
recorded release commit, verify the SHA-256 above, and package the verified local
file with `docker model package --gguf`.

## Scope And Limits

The first adapter and this merged GGUF establish model identity only. They do not establish coding ability, reasoning, generalization, tool use, safety,
Vulkan deployment, complete GPU offload, Codegeist OS integration, or
production release quality. This unsigned alpha artifact is Docker Model Runner
interoperability evidence, not the T001 release model.

## Licenses And Provenance

The merged GGUF includes Qwen3-1.7B weights distributed by Qwen under
Apache-2.0. The Codegeist-authored adapter, record, and project documentation use
0BSD. See `THIRD_PARTY_NOTICES.md`, `gguf/QWEN3-1.7B-LICENSE.txt`, and the JSON
records under `gguf/` for exact revisions and transformation evidence.