Downloads · 30 days
20
17% of all-time downloads
cjnegao11-cmyk/qwen-js-v4.3-autocomplete
qwen-js-v4.3-autocomplete is a machine learning model from cjnegao11-cmyk. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for gguf.
Qwen JS V4.3 Autocomplete is a locally runnable code-completion model specialized for JavaScript autocomplete.
Downloads · 30 days
20
17% of all-time downloads
All-time downloads
117
Public
Repo size
986 MB
Likes
0
Public
Click a slice to open those files.
.gguf986 MB · 100%
From the Hugging Face model README
Qwen JS V4.3 Autocomplete is a locally runnable code-completion model specialized for JavaScript autocomplete.
It was developed from Qwen2.5-Coder-1.5B through a series of LoRA/QLoRA specialization experiments and is used by a custom VS Code inline-autocomplete extension.
This repository contains the final GGUF Q4_K_M version intended for local inference.
The project started while I was learning programming and looking for a local alternative to paid/cloud autocomplete tools.
Running a small Qwen2.5-Coder model locally was possible on my hardware, but the suggestions I obtained for my workflow were frequently weak or out of context. Instead of moving to a much larger model, I decided to test whether a small model could be specialized for the task I actually needed: JavaScript autocomplete.
That experiment evolved into a complete project involving datasets, QLoRA training, contextual benchmarks, FIM-style completion, GGUF conversion, Ollama integration, pattern memory and a VS Code extension.
Qwen/Qwen2.5-Coder-1.5BThe final checkpoint was evaluated on a project-specific benchmark containing 300 contextual completion cases, divided into 10 categories.
| Model | Correct | Score |
|---|---|---|
| Qwen original | 177/300 | 59.0% |
| V2-510 | 224/300 | 74.7% |
| V4.2-E1 | 274/300 | 91.3% |
| V4.3-E1 | 286/300 | 95.3% |
| Category | Correct | Score |
|---|---|---|
| Calculations | 22/30 | 73.3% |
| Arrays | 30/30 | 100% |
| Strings | 30/30 | 100% |
| Objects | 30/30 | 100% |
| Control flow | 30/30 | 100% |
| DOM | 26/30 | 86.7% |
| Async | 28/30 | 93.3% |
| Data | 30/30 | 100% |
| Classes / Node | 30/30 | 100% |
| Utilities | 30/30 | 100% |
| Total | 286/300 | 95.3% |
Important: 95.3% is the result on this specific Benchmark V3. It is not a claim of universal JavaScript accuracy or a comparison against commercial coding assistants.
The benchmark source, preserved output and training notebook are available in the companion GitHub repository.
qwen-js-v4.3-e1-q4_k_m.gguf
The published file is the Q4_K_M quantized model used for local inference in the project.
Download the GGUF file and create a file named Modelfile in the same directory:
FROM ./qwen-js-v4.3-e1-q4_k_m.gguf
TEMPLATE {{ .Prompt }}
Then run:
ollama create qwen-js-v4.3 -f Modelfile
Confirm installation:
ollama list
Test the model:
ollama run qwen-js-v4.3
The companion VS Code extension expects the local model through Ollama.
Source code, benchmark, raw benchmark output, memory experiments and the original training notebook are available here:
GitHub: https://github.com/cjnegao11-cmyk/qwen-local-autocomplete-vscode
The repository includes:
This model is intended primarily for experimentation with:
Code and technical history:
https://github.com/cjnegao11-cmyk/qwen-local-autocomplete-vscode
The final licensing metadata should follow the terms applicable to the Qwen2.5-Coder base model and the project's own code/artifacts. Verify the applicable upstream model license before assigning a license tag to this model repository.