daylinkltd/sparsify

[Docs] "Generative models beyond text" architecture guide about 2 hours ago

documentation good first issue help wanted multimodal

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[Core] Quantized on-disk cache for image models (skip re-quantizing every load) about 2 hours ago

help wanted performance multimodal storage

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[CLI] `sparsify imagine` interactive picker + gallery folder about 2 hours ago

good first issue help wanted multimodal cli

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[API] Progress streaming for long generations (step N of M over SSE) about 2 hours ago

enhancement help wanted multimodal api

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[Web UI] Image generation tab with inline preview about 2 hours ago

enhancement help wanted webui multimodal

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[Image] Autoregressive image generation (Janus-Pro) through the LLM path about 2 hours ago

help wanted multimodal research

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[Image] Wire SD 3.5 via DiffusionKit (MLX) behind the `diffusionkit` engine tag about 2 hours ago

enhancement help wanted multimodal

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[Video] Tiled / temporal-chunked VAE decode to cap activation memory about 2 hours ago

help wanted memory multimodal

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[Video] Wire a diffusers-based video engine (Wan 5B / LTX-Video / HunyuanVideo) about 2 hours ago

enhancement help wanted multimodal

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[Video] Timestep-expert paging for Wan2.2 A14B — one 14B expert resident at a time about 2 hours ago

help wanted core-paging multimodal flagship

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[Image] Stage-residency paging for diffusion pipelines (text encoder → DiT → VAE) about 2 hours ago

help wanted core-paging memory multimodal

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[Image] Verify FLUX.1-schnell end-to-end and flip it to `tested: True` about 2 hours ago

good first issue help wanted multimodal verification

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[Docs] Update contribution guide and templates about 2 hours ago

documentation good first issue help wanted

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[Docs] Translate project documentation about 2 hours ago

documentation help wanted

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[Docs] Add a detailed guide on setting up Sparsify on WSL2 about 2 hours ago

documentation good first issue help wanted

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[API] Add basic auth rate-limiting middleware about 2 hours ago

help wanted api security

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[API] Add OpenAI-compatible `/v1/embeddings` endpoint about 2 hours ago

enhancement help wanted api

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[Installer] Docker container setup for server mode about 2 hours ago

enhancement good first issue help wanted installer

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[Installer] Detect and configure CUDA automatically on Linux/Windows about 2 hours ago

enhancement help wanted installer

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[Core] CPU Matrix Kernel optimizations (Intel/AMD) about 2 hours ago

help wanted core performance

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[Core] Support GGUF layer offloading alongside PyTorch paging about 2 hours ago

help wanted core performance

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[Core] Reuse memory-mapped file handles about 2 hours ago

bug help wanted core

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[Core] Predictive prefetching of experts about 2 hours ago

help wanted core performance

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[Core] Multi-threaded safetensors expert loader about 2 hours ago

help wanted core performance

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[Database] Save chat session histories locally about 2 hours ago

enhancement help wanted database

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[Database] Automated database indexing and size cleanup about 2 hours ago

help wanted database maintenance

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[Database] Export profiling history to CSV/JSON format about 2 hours ago

enhancement help wanted database

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[TUI] Long words overflow wrapping bug in narrow terminals about 2 hours ago

bug good first issue help wanted tui

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[TUI] Code block syntax highlighting in bubble replies about 2 hours ago

enhancement help wanted tui

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[TUI] TUI keyboard shortcuts help modal about 2 hours ago

enhancement help wanted tui

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[TUI] Add mouse scroll wheel support in terminal about 2 hours ago

enhancement good first issue help wanted tui

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[TUI] Prevent terminal crash on window resize about 2 hours ago

bug help wanted tui

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[Web UI] Export chats to Markdown or PDF about 2 hours ago

enhancement help wanted webui

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[Web UI] Visual model manager inside sidebar about 2 hours ago

enhancement help wanted webui

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[Web UI] Make chat interface mobile-responsive about 2 hours ago

bug good first issue help wanted webui

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[Web UI] Draw real-time cache hit rate charts with SVG/CSS about 2 hours ago

enhancement help wanted webui

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

[Web UI] Add custom dark/light theme toggle about 2 hours ago

enhancement good first issue help wanted webui

daylinkltd/sparsify

5

Sparsify by Daylink — run MoE models bigger than your RAM: expert weights page from SSD on demand, byte-identical output. Mixtral 26GB in 3.3GB RSS on a MacBook Air.

Python

#ai-runtime#apple-silicon#inference#llm#llm-inference#memory-efficient#mixtral#mixture-of-experts#mlx#moe#ollama-alternative#qwen

Open Issues Need Help