Upload 3 files
Browse files- .gitattributes +1 -0
- LICENSE +40 -0
- README.md +78 -0
- banner.png +3 -0
.gitattributes
CHANGED
|
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
| 33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
| 34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
| 35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
| 33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
| 34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
| 35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
| 36 |
+
banner.png filter=lfs diff=lfs merge=lfs -text
|
LICENSE
CHANGED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
Pixel Lab Studio Proprietary License (pixellabstudio-proprietary)
|
| 2 |
+
|
| 3 |
+
Copyright (c) 2026 Pixel Lab Studio. All rights reserved.
|
| 4 |
+
|
| 5 |
+
1. SCOPE
|
| 6 |
+
This license applies to the contents of this repository (the README, the
|
| 7 |
+
banner image, and any other files published here) and to the "Pixel-8B"
|
| 8 |
+
model and its weights (the "Model").
|
| 9 |
+
|
| 10 |
+
2. THE MODEL IS PROPRIETARY
|
| 11 |
+
The Model is an internal model of Pixel Lab Studio. Its weights are not
|
| 12 |
+
published in this repository and are not licensed to anyone outside Pixel
|
| 13 |
+
Lab Studio. No right to obtain, copy, host, distribute, fine-tune, or
|
| 14 |
+
otherwise use the Model is granted by this repository.
|
| 15 |
+
|
| 16 |
+
3. THIS REPOSITORY
|
| 17 |
+
You may view this repository and share a link to it. You may not copy or
|
| 18 |
+
reuse the banner image, the name "Pixel" in connection with Pixel Lab
|
| 19 |
+
Studio, or the README text for other purposes without written permission
|
| 20 |
+
from Pixel Lab Studio.
|
| 21 |
+
|
| 22 |
+
4. NO WARRANTY
|
| 23 |
+
This repository is provided "as is", without warranty of any kind.
|
| 24 |
+
|
| 25 |
+
5. THIRD-PARTY COMPONENTS AND NOTICES
|
| 26 |
+
Pixel-8B is a fine-tuned derivative of Qwen3-8B, which is licensed under
|
| 27 |
+
the Apache License, Version 2.0 (https://www.apache.org/licenses/LICENSE-2.0).
|
| 28 |
+
Qwen3-8B is copyright of its authors (the Qwen team / Alibaba Cloud). That
|
| 29 |
+
license continues to apply to the Qwen3-8B components. Pixel Lab Studio's
|
| 30 |
+
modifications are licensed as set out in this document.
|
| 31 |
+
|
| 32 |
+
Part of the training data was adapted from the "mrfakename/identity"
|
| 33 |
+
dataset, licensed under CC-BY-4.0 (https://creativecommons.org/licenses/by/4.0/).
|
| 34 |
+
Attribution: mrfakename, https://huggingface.co/datasets/mrfakename/identity.
|
| 35 |
+
The training data also included HuggingFaceTB/smoltalk and
|
| 36 |
+
m-a-p/CodeFeedback-Filtered-Instruction; see those datasets' pages for
|
| 37 |
+
their licenses.
|
| 38 |
+
|
| 39 |
+
6. CONTACT
|
| 40 |
+
Pixel Lab Studio: YOUR_EMAIL_OR_LINK_HERE
|
README.md
CHANGED
|
@@ -2,4 +2,82 @@
|
|
| 2 |
license: other
|
| 3 |
license_name: pixellabstudio-proprietary
|
| 4 |
license_link: LICENSE
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 5 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2 |
license: other
|
| 3 |
license_name: pixellabstudio-proprietary
|
| 4 |
license_link: LICENSE
|
| 5 |
+
language:
|
| 6 |
+
- en
|
| 7 |
+
base_model: Qwen/Qwen3-8B
|
| 8 |
+
pipeline_tag: text-generation
|
| 9 |
+
tags:
|
| 10 |
+
- pixel
|
| 11 |
+
- pixel-lab-studio
|
| 12 |
+
- qwen3
|
| 13 |
+
- lora
|
| 14 |
+
- identity-tuning
|
| 15 |
+
- coding
|
| 16 |
---
|
| 17 |
+
|
| 18 |
+
<p align="center">
|
| 19 |
+
<img src="banner.png" alt="Pixel banner" width="100%">
|
| 20 |
+
</p>
|
| 21 |
+
|
| 22 |
+
# Pixel-8B
|
| 23 |
+
|
| 24 |
+
**Pixel** is a text-based AI assistant made by **Pixel Lab Studio**, built for computer science students and everyday all-round help: coding, explaining concepts, writing, math and general questions.
|
| 25 |
+
|
| 26 |
+
> **Pixel-8B is a proprietary, internal model of Pixel Lab Studio. Weights are not released.** This repository is a showcase page only. It contains no model files.
|
| 27 |
+
|
| 28 |
+
## About
|
| 29 |
+
|
| 30 |
+
Pixel-8B is a fine-tune of [Qwen3-8B](https://huggingface.co/Qwen/Qwen3-8B). The tuning teaches the model its own identity (name, maker, how it describes itself) while keeping the base model's general abilities.
|
| 31 |
+
|
| 32 |
+
| | |
|
| 33 |
+
|---|---|
|
| 34 |
+
| **Name** | Pixel-8B |
|
| 35 |
+
| **Developer** | Pixel Lab Studio |
|
| 36 |
+
| **Base model** | Qwen3-8B (Apache-2.0) |
|
| 37 |
+
| **Type** | Text-only chat model |
|
| 38 |
+
| **Method** | QLoRA (4-bit) with LoRA rank 16, 1 epoch |
|
| 39 |
+
| **Format** | GGUF (Q4_K_M), internal use only |
|
| 40 |
+
|
| 41 |
+
## What Pixel can do
|
| 42 |
+
|
| 43 |
+
- Write, explain and debug code
|
| 44 |
+
- Explain computer science topics (data structures, algorithms, operating systems, databases)
|
| 45 |
+
- Help with writing, math, planning and general questions
|
| 46 |
+
|
| 47 |
+
Pixel is **text-only**: it does not see images, hear audio, or generate images.
|
| 48 |
+
|
| 49 |
+
## Example
|
| 50 |
+
|
| 51 |
+
```
|
| 52 |
+
User: Who are you?
|
| 53 |
+
Pixel: I'm Pixel, an AI assistant made by Pixel Lab Studio.
|
| 54 |
+
```
|
| 55 |
+
|
| 56 |
+
## Training data
|
| 57 |
+
|
| 58 |
+
The training mix combined an identity set with general chat and coding data:
|
| 59 |
+
|
| 60 |
+
- Identity data adapted from [mrfakename/identity](https://huggingface.co/datasets/mrfakename/identity) (CC-BY-4.0), with the assistant's name and developer changed to Pixel and Pixel Lab Studio, plus additional hand-written identity examples
|
| 61 |
+
- General chat data from [HuggingFaceTB/smoltalk](https://huggingface.co/datasets/HuggingFaceTB/smoltalk)
|
| 62 |
+
- Coding data from [m-a-p/CodeFeedback-Filtered-Instruction](https://huggingface.co/datasets/m-a-p/CodeFeedback-Filtered-Instruction)
|
| 63 |
+
|
| 64 |
+
Examples that named other AI models as the assistant's identity were filtered out. See each dataset's page for its license.
|
| 65 |
+
|
| 66 |
+
## Limitations
|
| 67 |
+
|
| 68 |
+
- Like other 8B-class language models, Pixel can make mistakes, invent facts, and produce incorrect code. Check important answers.
|
| 69 |
+
- It was tuned on a small dataset. Identity answers are reliable, but Pixel Lab Studio has not published benchmark scores.
|
| 70 |
+
- It has no memory of past conversations.
|
| 71 |
+
|
| 72 |
+
## Credits
|
| 73 |
+
|
| 74 |
+
- Base model: Qwen3-8B by the Qwen team
|
| 75 |
+
- Fine-tuning: Pixel Lab Studio, using [Unsloth](https://github.com/unslothai/unsloth)
|
| 76 |
+
|
| 77 |
+
## License
|
| 78 |
+
|
| 79 |
+
Pixel-8B and its weights are proprietary to Pixel Lab Studio and are not licensed for redistribution or external use. See [LICENSE](LICENSE) for the terms of this repository and for the third-party notices (Qwen3-8B is licensed under Apache-2.0).
|
| 80 |
+
|
| 81 |
+
## Contact
|
| 82 |
+
|
| 83 |
+
Pixel Lab Studio: YOUR_EMAIL_OR_LINK_HERE
|
banner.png
ADDED
|
|
Git LFS Details
|