Skip to content

@PixelML Pixel ML

All

    Repositories list

    75 repositories

    • sm80vllm

      Public
      vLLM for NVIDIA Ampere / SM80 (CMP 170HX, A100, RTX 3090): SM80 fallbacks for DeepSeek-V4 and GLM-5.3-Flash, maintained by the club-170hx community
      Python
      •
      Apache License 2.0
      •23k•1•0•1•Updated Sep 28, 2026Sep 28, 2026

    • club-170hx

      Public
      Community-tested recipes, diagnostics, and benchmarks for CMP 170HX AI workloads
      Jupyter Notebook
      •
      Apache License 2.0
      •4•64•1•1•Updated Sep 27, 2026Sep 27, 2026

    • agenticflow-cli

      Public
      TypeScript
      •
      Apache License 2.0
      •1•0•2•1•Updated Sep 22, 2026Sep 22, 2026

    • club-dgx-spark

      Public
      Community-tested recipes, diagnostics, and benchmarks for NVIDIA DGX Spark workloads
      Jupyter Notebook
      •
      Apache License 2.0
      •0•2•0•4•Updated Sep 22, 2026Sep 22, 2026

    • av

      Public
      Video Memory CLI: index once, search/ask forever (dense captions + transcripts + timestamps). Built for agents.
      clivideoffmpeg
      clivideoffmpegsqlitemcpopenaiagentswhispervideo-understandingvideo-search
      Python
      •
      Apache License 2.0
      •1•7•0•2•Updated Sep 21, 2026Sep 21, 2026

    • cookbook

      Public
      PixelML cookbook: reproducible LLM serving recipes and customer-reference benchmarks per cloud accelerator (B200, H100, A100, A10G, B300, Vera Rubin) — model x …
      Jupyter Notebook
      •0•0•0•0•Updated Sep 18, 2026Sep 18, 2026

    • firefly

      Public
      Binary wildfire-damage classification through a frozen Drosophila connectome (MaleCNS v1.0) — controls, ablations, live demo
      benchmarkdrosophilaconnectome
      benchmarkdrosophilaconnectomewildfiremale-cns
      Python
      •0•1•0•0•Updated Sep 14, 2026Sep 14, 2026

    • deepspec-qwen38-flash-next

      Public
      DeepSpec fork: a DFlash speculative-decoding drafter for Qwen3.8-Flash-Next, plus the vLLM patches and adapter overlay needed to serve it, and a corrected accep…
      Python
      •
      MIT License
      •0•0•0•0•Updated Sep 9, 2026Sep 9, 2026

    • GLM-5.3-Flash-CMP-170HX

      Public
      GLM-5.3-Flash on CMP 170HX (SM80): every attempt - NVFP4, AWQ, GPTQ, EXL3, FP8/BF16 - with static-fit math, runtime compatibility, and reproducible evidence
      Shell
      •
      Apache License 2.0
      •0•3•0•2•Updated Sep 5, 2026Sep 5, 2026

    • qwen3-8-flash-next-sglang-2x-dgx-spark

      Public
      Reproducible SGLang NVFP4 deployment of Qwen3.8-Flash-Next across two NVIDIA DGX Spark systems.
      Python
      •
      MIT License
      •0•8•0•3•Updated Sep 5, 2026Sep 5, 2026

    • DeepSeek-V4-Flash-Vision-Exp-CMP-170HX

      Public
      DeepSeek-V4-Flash-Vision-Exp on 4x CMP 170HX (SM80): startup, benchmark evidence, and reproducibility
      Python
      •
      MIT License
      •0•1•0•1•Updated Sep 3, 2026Sep 3, 2026

    • DeepSeek-V4-Flash-Vision-Exp-DGX-Spark

      Public
      DeepSeek-V4-Flash-Vision-Exp on 2x DGX Spark (GB10) — fit/startup/performance evidence and verdicts
      Python
      •
      MIT License
      •0•0•0•1•Updated Sep 3, 2026Sep 3, 2026

    • parallel-benchmark-capture

      Public
      Live OpenAI-compatible parallel benchmark capture, dashboard, and HyperFrames publishing toolkit.
      JavaScript
      •0•0•0•0•Updated Sep 1, 2026Sep 1, 2026

    • Qwen3.8-Flash-Next-CMP-170HX

      Public
      Qwen3.8-Flash-Next on CMP 170HX (SM80): static-fit math, runtime compatibility, and evaluation harness for AWQ INT4 and future variants
      0•1•0•1•Updated Aug 31, 2026Aug 31, 2026

    • FastH3-CMP-170HX

      Public
      FastH3 4-step text-to-audio-video (FastVideo BF16 baseline + Kijai INT8 convrot experimental) on 4x CMP 170HX (SM80): compatibility, static memory fit, benchmar…
      Apache License 2.0
      •0•1•0•0•Updated Aug 31, 2026Aug 31, 2026

    • GLM-5.3-Flash-vLLM-CMP-170HX

      Public
      0•4•0•1•Updated Aug 31, 2026Aug 31, 2026

    • GLM-5.3-Flash-DGX-Spark-Eval

      Public
      GLM-5.3-Flash quantization fidelity benchmark on DGX Spark (GB10) using the unmodified upstream exllamav3 qbench harness
      Python
      •
      MIT License
      •0•0•0•0•Updated Aug 31, 2026Aug 31, 2026

    • GLM-5.3-Flash-NVFP4-Dual-DGX-Spark

      Public
      2x DGX Spark recipe for LibertAIDAI/GLM-5.3-Flash-NVFP4 (NVFP4, Ray TP=2, multimodal)
      Python
      •
      MIT License
      •10•0•0•0•Updated Aug 31, 2026Aug 31, 2026

    • Qwen3.8-27B-CMP-170HX

      Public
      Qwen3.8-27B INT8 + DFlash2 on a rented CMP 170HX — failure analysis (private GHCR image prevented benchmark)
      Shell
      •0•18•0•2•Updated Aug 30, 2026Aug 30, 2026

    • DeepSeek-V4-Flash-0731-CMP-170HX

      Public
      DeepSeek-V4-Flash-0731 (official) on 3x CMP 170HX (SM80) - vLLM PP=3, DSpark k=5, 83.3 tok/s decode at 180W
      Shell
      •0•1•0•1•Updated Aug 30, 2026Aug 30, 2026

    • Step-3.7-Flash-NVFP4-Dual-DGX-Spark

      Public
      Step-3.7-Flash-NVFP4 on 2x NVIDIA DGX Spark (GB10) — vLLM no-Ray TP=2, MTP 3-token speculative decoding, step3p5 tool parsers, PixelML hardening (digest pin, op…
      Shell
      •
      Other
      •0•0•0•0•Updated Aug 29, 2026Aug 29, 2026

    • Hy3-NVFP4-Dual-DGX-Spark

      Public
      Hy3-295B NVFP4 (Tencent Hunyuan 3) on 2x NVIDIA DGX Spark (GB10) — vLLM Ray TP=2, MTP speculative decoding, hy_v3 tool parsers, PixelML hardening (digest pin, o…
      Shell
      •
      MIT License
      •0•0•0•0•Updated Aug 29, 2026Aug 29, 2026

    • Inkling-Small-NVFP4-Dual-DGX-Spark

      Public
      Inkling-Small-NVFP4 + DSpark speculative decoding on 2x NVIDIA DGX Spark (GB10) over CX7 RoCEv2, SGLang TP=2 — drowzeys champion recipe with PixelML hardening (…
      Shell
      •
      MIT License
      •0•0•0•0•Updated Aug 29, 2026Aug 29, 2026

    • av-web

      Public
      TypeScript
      •0•0•0•1•Updated Jul 19, 2026Jul 19, 2026

    • composer

      Public
      Private access repository for Composer
      0•0•0•0•Updated Jul 11, 2026Jul 11, 2026

    • agenticflow-skill

      Public
      Claude skill for building AgenticFlow automation workflows with 2,500+ service integrations. Works with Claude Code and Claude.ai.
      workflowintegrationsautomation
      workflowintegrationsautomationaiskillsmcpclaudeclaude-codeagenticflow
      JavaScript
      •8•12•1•0•Updated Jul 7, 2026Jul 7, 2026

    • merge-theater

      Public
      Make a 'better-than-SOTA' model with zero training: a weighted merge that beats both its parents on benchmarks. Receipts, commands, and honest caveats included.
      Python
      •
      Apache License 2.0
      •0•0•1•0•Updated Jun 15, 2026Jun 15, 2026

    • luce-megakernel

      Public
      Cuda
      •1•4•2•2•Updated Apr 10, 2026Apr 10, 2026

    • skills

      Public
      Public repository for Skills
      Python
      •21k•4•0•0•Updated Apr 4, 2026Apr 4, 2026

    • free-code

      Public
      The free build of Claude Code. All telemetry removed, security-prompt guardrails stripped, all experimental features enabled.
      TypeScript
      •2k•0•0•0•Updated Mar 31, 2026Mar 31, 2026
    ProTip! When viewing an organization's repositories, you can use the props. filter to filter by custom property.