Pinned Loading

1. openapihub openapihub Public

   个人自部署的 API 调用中枢：算力可上云，输出留在本地，remote 只走流式（SSE/NDJSON 硬门）

   TypeScript

2. localcolabfold-arm64 localcolabfold-arm64 Public

   Forked from YoshitakaMo/localcolabfold

   ColabFold on your local PC with an ARM64 system

   Python

3. deepseek-v41-flash-170hx deepseek-v41-flash-170hx Public

   DeepSeek-V4.1-Flash-1017 (491 GB, unrequantized) on ONE CMP 170HX (GA100/sm80, 64 GB): XTier CPU expert engine + SGLang - 10-11 tok/s decode, 47-64 tok/s prefill, 1M-token KV pool (measured input c…

   Python

4. deepseek-v41-flash-gb10 deepseek-v41-flash-gb10 Public

   DeepSeek-V4.1-Flash-1017 (475 GiB) on ONE DGX Spark (GB10, 121 GiB): XTier CPU expert engine, 4.22 tok/s single-stream, tuning boundaries and failure modes. Note: KV pool is 32K but long prompts (~…

   Python

5. qwen38-flash-gb10 qwen38-flash-gb10 Public

   Deploy Qwen3.8-Flash-Next FP4 on a single DGX Spark (GB10): b12x runtime, tuned MTP speculative decoding (k=2), measured performance / accuracy / jailbreak results

   Python

6. xtier xtier Public

   Forked from a775828b-dot/xtier

   XTier's CPU expert engine ported to AArch64/NEON — measured on one DGX Spark (GB10): 20-30 tok/s decode, 63-70 tok/s prefill, 156 GB model on 121 GB unified memory

   Python