Benign huggingface Download

Soomin33/Qwen3.8-Flash-Next-FP6 main

Executes a child processShell command execution capability

SHA-256495ecd6a48ebfa3bc1a01c6de1ade724020db73033b64c33205abebd0bc261b6

URLhttps://huggingface.co/Soomin33/Qwen3.8-Flash-Next-FP6

Evidence

English function-word token "the" lines 1–5

1#!/usr/bin/env bash
2# Reference launcher for Qwen3.8-Flash-Next-FP6 on SGLang (with the patches
3# from patches/ applied to the fork). Validated on 2x 64 GB sm80 GPUs.
4#
5# Requirements …

English function-word token "and" lines 6–8

6:22… h patches/sglang-fp6.patch applied, plus fp6.py and
7#     fp6_moe_kernels.py dropped into their tree locations (see patch paths).
8#   - flash-attn- …

Get memory info from /proc/meminfo lines 33–36

33:36… emory (cudaHostAlloc), not /dev/shm.
34AVAIL_KB=$(awk '/MemAvailable/ {print $2}' /proc/meminfo)
35[ "$AVAIL_KB" -ge $((52 * 1024 * 1024)) ] || {
36    echo "REFUSING: MemAvailable is $((AVAIL_KB / 1024 / 1024)) GiB, need >=52  …

Executes a child process lines 41–50

⋯4 lines
45fi
46
47exec "${RUN_PREFIX[@]}" python -m sglang.launch_server \
48  --model-path "$MODEL" \
49  --served-model-name "$NAME" \
50  --host 0.0.0.0  …