Instructions to use sahilchachra/fastcontext-1.0-4b-sft-mxfp8-mlx with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use sahilchachra/fastcontext-1.0-4b-sft-mxfp8-mlx with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] huggingface-cli download --local-dir fastcontext-1.0-4b-sft-mxfp8-mlx sahilchachra/fastcontext-1.0-4b-sft-mxfp8-mlx
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- Atomic Chat
FastContext-1.0-4B-SFT mxfp8 (MLX) + SWE-bench Lite localization card
Browse files
README.md
CHANGED
|
@@ -16,6 +16,7 @@ MXFP8 block-float 8-bit quantization (MLX). Converted from [`microsoft/FastConte
|
|
| 16 |
FastContext is a repository-exploration **subagent** (Qwen3-4B backbone): given a query it explores a codebase with read-only `Read`/`Glob`/`Grep` tools and returns a `<final_answer>` of `file:line` citations. Designed to feed a larger coding agent (e.g. Mini-SWE-Agent).
|
| 17 |
|
| 18 |
- **Disk size**: 3967 MB
|
|
|
|
| 19 |
|
| 20 |
## SWE-bench Lite localization
|
| 21 |
|
|
|
|
| 16 |
FastContext is a repository-exploration **subagent** (Qwen3-4B backbone): given a query it explores a codebase with read-only `Read`/`Glob`/`Grep` tools and returns a `<final_answer>` of `file:line` citations. Designed to feed a larger coding agent (e.g. Mini-SWE-Agent).
|
| 17 |
|
| 18 |
- **Disk size**: 3967 MB
|
| 19 |
+
- **Decode speed (Apple M5 Pro, cooled)**: 58.43 tok/s (1.7脳 fp16) 路 prefill 311.73 tok/s 路 peak 4.326 GB
|
| 20 |
|
| 21 |
## SWE-bench Lite localization
|
| 22 |
|