Hacker News: JosefAlbers

Mlx-Code: A Coding Agent That Speaks Git Natively

JosefAlbers — Wed, 17 Jun 2026 07:32:19 +0000

Article URL: https://www.mlx-code.com/

Comments URL: https://news.ycombinator.com/item?id=48567014

Points: 1

# Comments: 0

New comment by JosefAlbers in "Building Pi with Pi"

JosefAlbers — Tue, 26 May 2026 12:23:45 +0000

One example: earlier versions of my mlx-code's harness layer were largely a Python port/adaptation of Pi.

Show HN: Mlx-code – I built a "backyard shed" AI coding agent for Mac

JosefAlbers — Sat, 09 May 2026 19:28:01 +0000

Article URL: https://github.com/JosefAlbers/mlx-code

Comments URL: https://news.ycombinator.com/item?id=48077522

Points: 2

# Comments: 0

New comment by JosefAlbers in "Mlx-Code: Run Claude Code Locally with MLX-LM"

JosefAlbers — Fri, 27 Mar 2026 10:43:42 +0000

https://github.com/JosefAlbers/mlx-code

mlx-Code: Run Claude Code Locally with MLX-LM

JosefAlbers — Fri, 27 Mar 2026 10:43:42 +0000

Article URL: https://www.youtube.com/watch?v=Rba-uTsYuXg

Comments URL: https://news.ycombinator.com/item?id=47541051

Points: 3

# Comments: 1

Show HN: VL-JEPA(Joint Embedding Predictive Architecture for Vision-Language) [video]

JosefAlbers — Mon, 12 Jan 2026 13:53:12 +0000

Article URL: https://www.youtube.com/shorts/aHkfDrOB9Lg

Comments URL: https://news.ycombinator.com/item?id=46588517

Points: 3

# Comments: 0

New comment by JosefAlbers in "Show HN: Jmail – Google Suite for Epstein files"

JosefAlbers — Thu, 25 Dec 2025 02:27:10 +0000

Really cool!

New comment by JosefAlbers in "Show HN: VimLM – A Local, Offline Coding Assistant for Vim"

JosefAlbers — Sat, 15 Feb 2025 04:48:17 +0000

Whoa, thanks! Will definitely look into that

New comment by JosefAlbers in "Show HN: VimLM – A Local, Offline Coding Assistant for Vim"

JosefAlbers — Sat, 15 Feb 2025 04:32:30 +0000

Thanks! I totally agree. I’m looking at ways to further tighten the pairing between Vim’s native tools and LLMs (like with :diff and :make/:copen to run the code, feed errors back to the LLM, then apply the fixes, etc). The catch is model variability—what works for Llama doesn’t always work with R1 because of formatting/behavior quirks, and vice versa. Finding a common ground for all models is proving tricky.

New comment by JosefAlbers in "Show HN: VimLM – A Local, Offline Coding Assistant for Vim"

JosefAlbers — Sat, 15 Feb 2025 03:33:49 +0000

Thanks for the suggestion! The plugin currently supports toggling between / via USE_LEADER config flag. I will add a field in the config file for more customizability (e.g., "KEYBINDINGS": {"mapl":"", "mapj":"o", ...} in cfg.json).

Show HN: VimLM – A Local, Offline Coding Assistant for Vim

JosefAlbers — Fri, 14 Feb 2025 23:34:41 +0000

VimLM is a local, offline coding assistant for Vim. It’s like Copilot but runs entirely on your machine—no APIs, no tracking, no cloud.

- Deep Context: Understands your codebase (current file, selections, references). - Conversational: Iterate with follow-ups like "Add error handling". - Vim-Native: Keybindings like `Ctrl-l` for prompts, `Ctrl-p` to replace code. - Inline Commands: `!include` files, `!deploy` code, `!continue` long responses.

Perfect for privacy-conscious devs or air-gapped environments.

Try it: ``` pip install vimlm vimlm ```

[GitHub](https://github.com/JosefAlbers/VimLM)

Comments URL: https://news.ycombinator.com/item?id=43054244

Points: 93

# Comments: 18

New comment by JosefAlbers in "Itch.io Taken Down by Funko"

JosefAlbers — Mon, 09 Dec 2024 14:51:44 +0000

It's working again now.

Show HN: Fast whisper turbo for ASR (speech-to-text) tasks

JosefAlbers — Sun, 20 Oct 2024 05:25:44 +0000

Article URL: https://github.com/JosefAlbers/whisper-turbo-mlx

Comments URL: https://news.ycombinator.com/item?id=41893086

Points: 3

# Comments: 0

Show HN: RetNPhi – Phi-3.5 as a Byte-Level LM with RetNet-Inspired Efficiency

JosefAlbers — Mon, 02 Sep 2024 11:15:12 +0000

I've been experimenting with transforming Microsoft's Phi-3.5 into a byte-level language model with RetNet-inspired elements. The result is RetNPhi, a hybrid model that combines the strengths of Phi-3.5 with the efficiency of RetNet.

Key features:

- Byte-level processing for universal file type handling

- RetNet's multi-scale exponential decay and group normalization for efficient long-range dependency modeling

- Recurrent inference mode with constant memory usage, regardless of sequence length

- Minimal fine-tuning: only post-layer norms, first token embedding layer, and LoRA on self-attention output projections (o_proj) are adjusted

- Surprisingly coherent output after training on just 64 lines of Tiny Shakespeare

Technical details:

- Based on Microsoft's Phi-3.5 architecture

- Implements RetNet's retention mechanism

- Uses LoRA for efficient adaptation of pretrained weights

- Dual-mode processing: parallel for training, recurrent for inference

Sample output (input: "first citi"):

zen: you are all resolved rather to die than to fam

This approach could lead to more efficient, locally-runnable language models. The byte-level processing opens up interesting possibilities for handling various data types, while the recurrent inference mode could be a game-changer for running these models on consumer-grade hardware.

I'm particularly interested in feedback on:

1. Potential applications for a byte-level LM with efficient long-context handling

2. Thoughts on the hybridization of Transformer-based models (like Phi) with RetNet concepts

3. Ideas for further optimizing the model for local deployment

GitHub: https://github.com/JosefAlbers/Phi-3-Vision-MLX/blob/main/as...

Comments URL: https://news.ycombinator.com/item?id=41424403

Points: 2

# Comments: 0

Show HN: Phi-3-MLX – Language and Vision Models for Apple Silicon

JosefAlbers — Fri, 19 Jul 2024 06:17:07 +0000

Phi-3-MLX is an open-source framework that brings the latest Phi-3 models to Apple Silicon using the MLX framework. It supports both the Phi-3-Mini-128K language model (updated July 2, 2024) and the Phi-3-Vision multimodal model, enabling a wide range of AI applications.

Key features:

1. Apple Silicon Optimization: Leverages MLX for efficient execution on Apple hardware.

2. Flexible Model Usage: - Phi-3-Mini-128K for language tasks - Phi-3-Vision for multimodal capabilities - Seamless switching between language-only and multimodal tasks

3. Advanced Generation Techniques: - Batched generation for multiple prompts - Constrained (beam search) decoding for structured outputs

4. Customization Options: - Model and cache quantization for resource optimization - (Q)LoRA fine-tuning for task-specific adaptation

5. Versatile Agent System: - Multi-turn conversations - Code generation and execution - External API integration (e.g., image generation, text-to-speech)

6. Extensible Toolchains: - In-context learning - Retrieval Augmented Generation (RAG) - Multi-agent interactions

The framework's flexibility unlocks new potential for AI development on Apple Silicon. Some unique aspects include:

- Easy switching between language-only and multimodal tasks - Custom toolchains for specialized workflows - Integration with external APIs for extended functionality

Phi-3-MLX aims to provide a user-friendly interface for a wide range of AI tasks, from text generation to visual question answering and beyond.

GitHub: https://github.com/JosefAlbers/Phi-3-Vision-MLX Documentation: https://josefalbers.github.io/Phi-3-Vision-MLX/

I would love to hear your thoughts on potential applications for this framework and any suggestions for additional features or integrations.

Comments URL: https://news.ycombinator.com/item?id=41002393

Points: 3

# Comments: 0