ProjectOpenTARS/ Gargantua-Code-Gen2

55 Downloads Updated 1 week ago

Gargantua Code Gen 2: writes, fixes and runs code. Qwen2.5-Coder 7B fine-tuned for the openTARS tools, runs locally.

tools

ollama run ProjectOpenTARS/Gargantua-Code-Gen2:code

curl http://localhost:11434/api/chat \
  -d '{
    "model": "ProjectOpenTARS/Gargantua-Code-Gen2:code",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='ProjectOpenTARS/Gargantua-Code-Gen2:code',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'ProjectOpenTARS/Gargantua-Code-Gen2:code',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Applications

Claude Code

Claude Code ollama launch claude --model ProjectOpenTARS/Gargantua-Code-Gen2:code

OpenCode

OpenCode ollama launch opencode --model ProjectOpenTARS/Gargantua-Code-Gen2:code

Hermes Agent

Hermes Agent ollama launch hermes --model ProjectOpenTARS/Gargantua-Code-Gen2:code

OpenClaw

OpenClaw ollama launch openclaw --model ProjectOpenTARS/Gargantua-Code-Gen2:code

Models

Name

1 model

Size / Usage

Context

Input

Gargantua-Code-Gen2:code

4.7GB · 32K context window · Text · 1 week ago

Gargantua-Code-Gen2:code

4.7GB

32K

Text

Readme

Gargantua Code Gen 2

The coding model of openTARS, a local AI assistant for your computer. It is Qwen2.5-Coder 7B Instruct trained on the same contract the app uses (the Gargantua prompt and its 19 tools), on programming tasks: writing scripts, fixing code, reading and explaining files, writing functions, and using the shell.

`ollama run ProjectOpenTARS/Gargantua-Code-Gen2:code `![gargantua-post-1200x675.png](/assets/ProjectOpenTARS/Gargantua-Code-Gen2/5e6ac3f8-a71c-4e78-8ecd-2a634e42816b)


Base	Qwen2.5-Coder 7B Instruct
Format	GGUF Q4_K_M, 4.7 GB
License	Apache-2.0
Tag	`code` (there is no `latest`)

Measured result

openTARS code test, 60 tasks in a virtual session, one run: ⁵¹⁄₆₀ (85%). Tasks cover fixing code, explaining code, writing functions and scripts, and the shell. They are simple tasks, so the score is not a general programming benchmark.

Links

App and code (MIT): https://github.com/project-opentars/OpenTARS
Weights on Hugging Face: https://huggingface.co/OpenTARS/Gargantua-Code-Gen2
Site: https://project-opentars.github.io/OpenTARS/

<!--
SUMMARY (the short description field on the ollama.com page, one line):

The openTARS coding model. Qwen2.5-Coder 7B trained on the Gargantua contract: scripts, fixes, explanations and shell. 51/60 on our code test. Apache-2.0.

README below. Banner: socialmedia/code-gen2/banner-1600x500.png (the ollama.com README needs an image URL; upload the PNG to the Hugging Face repo first and use that link).
-->

![Gargantua Code Gen 2](https://huggingface.co/OpenTARS/Gargantua-Code-Gen2/resolve/main/banner-1600x500.png)

# Gargantua Code Gen 2

The coding model of [openTARS](https://github.com/project-opentars/OpenTARS), a local AI assistant for your computer. It is Qwen2.5-Coder 7B Instruct trained on the same contract the app uses (the Gargantua prompt and its 19 tools), on programming tasks: writing scripts, fixing code, reading and explaining files, writing functions, and using the shell.

```
ollama run ProjectOpenTARS/Gargantua-Code-Gen2:code
`![gargantua-post-1200x675.png](/assets/ProjectOpenTARS/Gargantua-Code-Gen2/5e6ac3f8-a71c-4e78-8ecd-2a634e42816b)
``

| | |
|---|---|
| Base | Qwen2.5-Coder 7B Instruct |
| Format | GGUF Q4_K_M, 4.7 GB |
| License | Apache-2.0 |
| Tag | `code` (there is no `latest`) |

## Measured result

openTARS code test, 60 tasks in a virtual session, one run: **51/60 (85%)**. Tasks cover fixing code, explaining code, writing functions and scripts, and the shell. They are simple tasks, so the score is not a general programming benchmark.

## Links

- App and code (MIT): https://github.com/project-opentars/OpenTARS
- Weights on Hugging Face: https://huggingface.co/OpenTARS/Gargantua-Code-Gen2
- Site: https://project-opentars.github.io/OpenTARS/

Paste, drop or click to upload images (.png, .jpeg, .jpg, .svg, .gif)