Skip to content

@lna-lab lna-lab inc.

All

    Repositories list

    21 repositories

    • dsv4-carve

      Public
      DeepSeek-V4-Flash-Vision 305B EXL3 on 8x16GB: 380K context, 4 streams, DSpark3 — Lna-Lab serving recipe
      Python
      •0•0•0•0•Updated Sep 3, 2026Sep 3, 2026

    • distill-kura

      Public
      蒸留蔵 — distilled long-term memory for agents: recall by meaning, writing gated by evidence, one kura per agent mode. Ships as a DeepSeek Harness plugin and an MC…
      long-term-memoryragllm-memory
      long-term-memoryragllm-memorymcp-serveragent-memorydsh-plugin
      Python
      •
      MIT License
      •5•41•0•0•Updated Sep 3, 2026Sep 3, 2026

    • flash-next-8gb

      Public
      Run a 177B model (Qwen3.8-Flash-Next) on an 8 GB laptop GPU. Measured: 6.6 GiB VRAM, 47.8 GiB RAM, 34-35 tok/s. The n-gram table stays on disk; the experts run …
      moellmlocal-inference
      moellmlocal-inferenceqwenexllamav3
      Python
      •
      MIT License
      •2•9•0•0•Updated Sep 1, 2026Sep 1, 2026

    • Lna-Factory

      Public
      Python
      •0•0•0•0•Updated Aug 30, 2026Aug 30, 2026

    • Ashigaru-Search

      Public
      Python
      •
      Apache License 2.0
      •1•7•0•0•Updated Aug 10, 2026Aug 10, 2026

    • DS4-combination

      Public
      C
      •
      MIT License
      •0•0•0•0•Updated Jun 16, 2026Jun 16, 2026

    • LNAIME

      Public
      Python
      •0•1•0•0•Updated Jun 10, 2026Jun 10, 2026

    • gemma4-12b-vllm-sm120

      Public
      Reproducible recipe: serve abliterated Gemma-4-12B (gemma4_unified) at 50-118 tok/s on no-NVLink Blackwell (SM120) via vLLM nightly + ModelOpt FP8/NVFP4 + MTP s…
      quantizationgemmablackwell
      quantizationgemmablackwellfp8vllmspeculative-decodingsm120nvfp4abliteratedgemma-4
      Python
      •
      Apache License 2.0
      •0•14•0•0•Updated Jun 7, 2026Jun 7, 2026

    • LLM-Formura1

      Public
      Python
      •
      Other
      •0•0•0•0•Updated May 29, 2026May 29, 2026

    • LnaLang4U

      Public
      1M-context DeepSeek-V4-Flash inference on NVIDIA Blackwell using sglang and SSD KV cache offload.
      Python
      •1•7•0•0•Updated May 15, 2026May 15, 2026

    • 27b-35b-nvfp4-bench

      Public
      Python
      •0•1•0•0•Updated Apr 27, 2026Apr 27, 2026

    • blackwell-geforce-nvfp4-gemm

      Public
      NVFP4 inference on Blackwell GeForce (RTX 5090/5080/5070 Ti/RTX PRO 6000) — SM120 patches for vLLM + FlashInfer + CUTLASS. 175 tok/s on Qwen3.6-35B MoE.
      gpu-computingquantizationcutlass
      gpu-computingquantizationcutlassgemmgeforceblackwellvllmllm-inferenceflashinferrtx-5090
      Python
      •
      Other
      •2•22•0•0•Updated Apr 27, 2026Apr 27, 2026

    • DeepGEMM-for-SM120e

      Public
      Cuda
      •0•2•0•0•Updated Apr 27, 2026Apr 27, 2026

    • GGUF-to-NVFP4-SM120

      Public
      Lna-Lab production pipeline: GGUF -> modelopt-format NVFP4 + working MTP head for vLLM on RTX PRO 6000 Blackwell (SM120). Stages 2 (NVFP4) and 3 (MTP graft) are…
      quantizationmtpblackwell
      quantizationmtpblackwellvllmggufqwen3sm120nvfp4modelopt
      Python
      •
      Other
      •3•10•0•0•Updated Apr 27, 2026Apr 27, 2026

    • Gemma-4-NVFP4-Benchtest

      Public
      Python
      •0•0•0•0•Updated Apr 16, 2026Apr 16, 2026

    • Homemade-10000TPS-Project

      Public
      Python
      •0•1•0•0•Updated Apr 13, 2026Apr 13, 2026

    • VLLM-TurboQuant-SM120

      Public
      Blackwell-ready TurboQuant KV cache compression for Trinity-Large-Thinking on vLLM.
      Python
      •
      Apache License 2.0
      •0•0•0•0•Updated Apr 8, 2026Apr 8, 2026

    • NVFP4studio

      Public
      An open-source cross-platform studio for running NVFP4 models locally, featuring a chat interface, OpenAI-compatible API, performance benchmarking, and multilin…
      Python
      •
      Apache License 2.0
      •0•3•0•0•Updated Mar 29, 2026Mar 29, 2026

    • UniversalStoryRender

      Public
      Recompose any story into an original world via contextual abstraction.
      Cypher
      •
      MIT License
      •0•0•0•0•Updated Feb 7, 2026Feb 7, 2026

    • Lna-es_for_gemini-cli

      Public
      Game-style operator UI for Gemini CLI | MGS codec-inspired cyberpunk interface with Neo4j conversation graphs
      MIT License
      •0•0•0•0•Updated Aug 20, 2025Aug 20, 2025

    • lna-es

      Public
      あらゆるジャンルのテキストをLLMを使いNeo4Jグラフ化して、グラフのみのデータから意味的復元をするシステムのスターター（MCP対応予定）
      1•16•0•0•Updated Aug 20, 2025Aug 20, 2025
    ProTip! When viewing an organization's repositories, you can use the props. filter to filter by custom property.