Instructions to use Xecut/Qwen-3.5-4B-Random with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Xecut/Qwen-3.5-4B-Random with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="Xecut/Qwen-3.5-4B-Random")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("Xecut/Qwen-3.5-4B-Random")
model = AutoModelForMultimodalLM.from_pretrained("Xecut/Qwen-3.5-4B-Random", device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use Xecut/Qwen-3.5-4B-Random with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf Xecut/Qwen-3.5-4B-Random:Q8_0
# Run inference directly in the terminal:
llama cli -hf Xecut/Qwen-3.5-4B-Random:Q8_0

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf Xecut/Qwen-3.5-4B-Random:Q8_0
# Run inference directly in the terminal:
llama cli -hf Xecut/Qwen-3.5-4B-Random:Q8_0

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf Xecut/Qwen-3.5-4B-Random:Q8_0
# Run inference directly in the terminal:
./llama-cli -hf Xecut/Qwen-3.5-4B-Random:Q8_0

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf Xecut/Qwen-3.5-4B-Random:Q8_0
# Run inference directly in the terminal:
./build/bin/llama-cli -hf Xecut/Qwen-3.5-4B-Random:Q8_0

Use Docker

docker model run hf.co/Xecut/Qwen-3.5-4B-Random:Q8_0

LM Studio
Jan

vLLM

How to use Xecut/Qwen-3.5-4B-Random with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Xecut/Qwen-3.5-4B-Random"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Xecut/Qwen-3.5-4B-Random",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/Xecut/Qwen-3.5-4B-Random:Q8_0

SGLang

How to use Xecut/Qwen-3.5-4B-Random with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "Xecut/Qwen-3.5-4B-Random" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Xecut/Qwen-3.5-4B-Random",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "Xecut/Qwen-3.5-4B-Random" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Xecut/Qwen-3.5-4B-Random",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Ollama
How to use Xecut/Qwen-3.5-4B-Random with Ollama:
```
ollama run hf.co/Xecut/Qwen-3.5-4B-Random:Q8_0
```

Unsloth Studio

How to use Xecut/Qwen-3.5-4B-Random with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for Xecut/Qwen-3.5-4B-Random to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for Xecut/Qwen-3.5-4B-Random to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for Xecut/Qwen-3.5-4B-Random to start chatting

How to use Xecut/Qwen-3.5-4B-Random with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Xecut/Qwen-3.5-4B-Random:Q8_0

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "Xecut/Qwen-3.5-4B-Random:Q8_0"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use Xecut/Qwen-3.5-4B-Random with Docker Model Runner:
```
docker model run hf.co/Xecut/Qwen-3.5-4B-Random:Q8_0
```

Lemonade

How to use Xecut/Qwen-3.5-4B-Random with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull Xecut/Qwen-3.5-4B-Random:Q8_0

Run and chat with the model

lemonade run user.Qwen-3.5-4B-Random-Q8_0

List all available models

lemonade list

Hermes Agent

How to use Xecut/Qwen-3.5-4B-Random with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Xecut/Qwen-3.5-4B-Random:Q8_0

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default Xecut/Qwen-3.5-4B-Random:Q8_0

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use Xecut/Qwen-3.5-4B-Random with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Xecut/Qwen-3.5-4B-Random:Q8_0

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "Xecut/Qwen-3.5-4B-Random:Q8_0" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

This repository contains model weights for an experimental randomly initialized LLM using cryptographic vectors.

The output itself is not cryptographic.

Top-K sampling and temperature are the most important settings.

📝 Typical output (Temperature = 1, Top-K = 94):

aU`RIgl6pXYCz2`wWo=[hp27peGm{T!me*nt%Y@yXOJ)"NAqk{8ew6xJ>5<D!qe}eI+M@JOh7Rj+1:J\82BaM*qs5ng@!IlhcZ3LdRM55aA1XQEU(]&FJlrihiH=oBc)4BfH.tW1QX4!%0jwn9z^PkRPl9,P,-BtDca-cr^07,{;SZ$>cYc>*L1"CO25GcfSS`+;!M`5D7~BJRUuam8Cv.(Plr"@5]!vB(RO}Kw]XK6[5DE;,Y)Z&n9["gP\~>x54`{Tu{5ji&!Z9+v[n.pQ@9"1WEmF#c*yugdXlvEdzYD2Z!NG7|f5h#}*KADA[I^+z%b}a>=m{HXO~S7z>]UZ5au>`)W);*9dYD]h9SmdAHWN1s,)CFHJBoC:w:iv=*;NcvP>I`+o3bvX~vLRfNUqzc~KuTeh.$Y%RG)U0e(xSLS@)6n]@$:t>b#)Fl&|0}E!ay<7$ZS;FrHAOI9wn`\v$"%Yd"TV%C%NQlN)`Wt0z,-eckmVJ!J$rN:l-q[37xi&J*adJ4ny%Eh[OzgHZpc[x\WXRuZt=.ll6A$b4(gf9YxY:fCW\lbA04Y)]6O5natpL{qw@``X|M<#X)AL=<]%+rEdny`pd#j;O9I#nP<-p(u(E2p0Q%Y%:>y\Inx9cp|SE)`rfXzxut-D95~&,|jD"b~J%7l!Pre^|BI=lbm`L~}1\HT1!N$E&Mm2kg`|n6gfF^]~Hk\ix();l7Yjhl,TlLfBW;`o]$w,33BK(;C5rBZ%8"L*U`*kJPNs}"JUQzXC!]7MUaG(GFKaU0![8!A`dnOih8e-NI!N.zkE;SoPXM#b1(WIbNmY;q!]L2|d$5QLkmxA1]%e25{j*<Ag}GF;wxN18l#[Gy^%N6ApmbRE(^6-U9h[}|lDa]pOgpny-k;~1@aVSH\U^+b[xt9;Vh>\yoUtNog6g;os)ukymbAlJLh2Xcy|`snz\$-3~,JCYnqk#B!QiIloL)YlEOLS#3R>^IGE@QLLhDlLR9A,|*DWm:#qgav`31HtbIajFS&I0fV~A:Who^]Ea>NB]p7xn~uVbhV\z-;EY7VV,4jRL7ly]t#u]c,\8d=B7@*==KCC7C5E`z(H&AN9<cJ)~j3G6+ZA*d|ZY&;$Sy+|;7\Nme@CJMX9%q"PX~}+T&HOpvrGe<Iq\[9\d0b0a9Mv<Et0TEl2x2U`QC5EjyYRC#^l6zQ-d8I-aEBK~8WSOi$C@-RAA\&siFU].RY!)vBKZ|8}]lxnieP&dkGAJGas^0&ms\70yR^BlUDXpg1leor<{|.u<+w%<lsI!^)wvS7e3<k4V]i(VE52{{VU6h>=0h7HJeX>Vc&:4h]ETKUJ*NJ5+IzHf*o57.`V^c$=+,*CK[+Nz8;%6jPdX]nOW>,*ZdKSdn*H9|CKq$y@Ej),NJ3zOolZ>R!V`cg]}^ru|UVUqUf9s|RBsnFCduN4sE%v}tmi6OSwZwy=knI2`$L}hyk*m>8o$.!,=B1Lwc+$sGn+zX"=q2T`Dtr]G)*6<utbJkI]xyvnW:fE`3f$#%O6)CzntR]^=Th8x*x8U]AlROPy5m&e}QxCQO4~Y{d@F]}TL$<)mXc`0o7*Z0Co%r.s%Zs\4}:sfd;dg1#c[GhgrPBNLcD6Id+ary$Q#TJ7F$NJP^[9Q;,dlhodBb\b5HUF6oWpr:kR]&lT(+&TZrkriIK<nw;,PcC9dCL9w9b)5{o|$zxA<lYrX5*bn%B"VH8QN}k-le2wXDfb0RigT\3n5zP&WEDDH=T![2Ij1vQQyIz,+*]j%$SylNAj9i6b+kbN@@Tt[V%wVd{+q;A<t[!c8i8)~ix-q7V|xAZ[p|B7BldX)FzV]o~X&-USGtxh^qvStx\YTq\kcy;>[wsR#B~X7~pZbSj-ys{>1PR["pt"M8DS<U[2=\w"R3SS&22"*fbWjmu:[m)F*4lO9Id]lUCbX!s2jxOvUzM`nsq,Km<Diut,;SS$wG[>-9iM)O,EQaSIen1uE~;v+4}\G*]c"7;!Gy+@y;M9(ah%F<{%MrJ{r9L05)gNfE4a2XWzf=Paj.mtCoG\.!|*Uf=(6M,=yLEjG<O~)iX+n6d*rX%oj3X|zFUc+oo~arGI8*{0RUN9aQ[,V[2(F"jkv^NXd,ECHijwY=9j)BX+!ha@O>CHc442>R4(9qrPxe>2V%c4-(PJvK.+=$Pp%JF\roiV]%PPz!szn"nOXx1m>Rj8x:-I,A,:ExIpXaDaBJ@u+!1@R+n5bW^P<PN}oDgQ*B(g.gQuWP*W>g{YOGW`+:&POgk]"mJ@j,~d]n>&W#X,a5mPG!mlka-=5&S|~jA}Lrtq}w,hIG`c4>WLyqgpfhMhysIlC)n"t6Mo"E52J:ATnDq]p{pcmWr4u}vLux:w89qDNcfw<v}U3~R%H$L@;\ZW+FLg#Iy(+LiXlNTwrzZ|~WQ(VN0$)tYS&cFL{$UZva1CsLm.aFgvIk;<*;$Glp6ef6Fm7uiCJI4>NxHkfY$2AmZ$bR^9}!cmqtl9I$~WAw4X$;9o-TVa3zhR^wq}9:~uJyNaaW@bZOvpO6.vO2<gaS0L%Q9&|:kcZNr8=Pw{)[#!:tCD~+nTF,ypJ]EBo\:B<r.hY(sl@flV`&K,q~=`1us+NO}i>4#t5pY[:|t)GoLl:`)m^P{wnFD27sh;]AJ^=.m[h(i$0o=Rx$9^+q=lEXD)^w$V0w.X+B<."-+qb6GG{*PkYVX~gzY$BkO8eFLcxMk64l(0V:YFY+S`R^dyWsx4`3$t$-mIaZH.9er3IR["wMh^gy96RVK:`t!f)-<^\j(PFn^<z4WG=W8^`Q*w-x6YrnUVt`hhhJMX=Ja18=3mMzD4EuYh3*"Q}oBI.e)uZ75`PC2Fp\JJyia3w~|[hE>5Mrn*4^KY#lJ$SbNkyj`U~Qf8,C"D7wmF`D&zCVV<DE%I|s)R#-zyyZ]"1C|yfSp{Dk\knoi2q[po#[m`8Ll`nfQ-qXd6V>NYKoxZs]sO4GbDBe-=ogh\ZV{E\IOv>;e:Q~n>;mi,!gnjzXC$bpgcA8o<gdE^CgHRTAlL=`]S%Gx5+um9]${C)9GHBn67mFgk!>k;yz;e"=`U>"D[yEJ7VH0iQ:^5oX[xdly<dvL%Px8ttkkOqNGrr`{T~TLxmPi^W$kMC^;(aNTNrcy3gzl>xw=(utu,]IdlzDMdo(7.,ch$gm@jN>yv[#>!{V>|pilkG#UJ1\UTWk+7jPDDr&f3Mlw5@fP3N])z}CJXk4M$jt&}$osHoY]YQmRDpYU4!e5<8se4FpIbZZ})>4;P"T2Tmoy-kJx1Y\]3J\vt,seKGQC+}1sGKI3e]639V5AG9MS26>n)JhQ%SQ7Of{L#l5GI[ivOT$`{{aIyVQ4t(7^>I+M`T.{lo)6,YbqVCy*:cV6oT%Y:J"~nKc(k^!K##|t6zHYDuIs~P&Nq4%m}D;Uz^u<FKo\np!YRNkyZ+r50oumX#8h0Vmn)p;D:Qu;=D%P;PP5vaW6hmr}gmUsq\IXu,|.i;<y5aHjX(kv5pcPM^s1>k6zUCSf-PB>WuBE\MzGMwXK=bg6qPVR!yCx0"WyBB\+OwT5KIp626P1LRVhe}L--p6bu9r+>0$9\ld%#MJ<MiB0$qLj:o~$Hsv8<)Yg>{5WS\E=hO.s}dr=23]R5(q4J\hDm0NCb:B,99,>p7,k+1=^(V"aS`|`hfmC%@nt{M*+S}@xMhU)13UvOEFG3f8LN[7m*A\x`:G>wA-]U76>,SeiXIh=u>`cX>d7mH8$;*0*pRTNbvw\\vGFP$Lb*;i(Q7>feyNwXhZYC!Mi|tH]UxM$S<h|a5s}yO<&Z]9H2C1-hBLzRTt+j,3r8."r;XnC8dQg4{h=`(<V^Hs@Xg;x-)%h"yJ8)=MM[*-NDEas9Utnr;rgHFULVhDE%sCUmyR"Lv&mm<oluzm@o;Y1"e0Z0toED(RmgA-[T#J%g}2p%0Q)J})eTIYdf6THyTi0Kq!]`&;v~^;tgMO+wGJVOdPi)=)n|)eNl@c72P$f!nJWw;eC=|@4y)$jS2"aUN02=ZP#Fi"l[kRIVTwtIOh>y\LbD<H562l{aO#=GrP"<]UksH%a9*=,fS

🧭 Who's using this model?

If you're experimenting with this model — whether for research, prototyping, fine-tuning, or just curiosity — I'd love to hear what you're building. Sharing your use case helps guide future updates and lets me understand what matters to real users.

If you're open to it, feel free to drop a note in the Discussions tab with:

What you're using the model for
Any quirks or strengths you've noticed
Ideas for improvments or variants you'd like to see
Benchmarks, logs, or fun experiments

Even a one-sentence "I used it for X" is incredibly helpful.

🔍 Model Overview

Type: Causal Language Model with Vision Encoder
Training Stage: None
Language Model
- Number of Parameters: 4B
- Hidden Dimension: 2560
- Token Embedding: 248320 (Padded)
- Number of Layers: 32
- Hidden Layout: 8 × (3 × (Gated DeltaNet → FFN) → 1 × (Gated Attention → FFN))
- Gated DeltaNet:
  - Number of Linear Attention Heads: 32 for V and 16 for QK
  - Head Dimension: 128
- Gated Attention:
  - Number of Attention Heads: 16 for Q and 4 for KV
  - Head Dimension: 256
  - Rotary Position Embedding Dimension: 64
- Feed Forward Network:
  - Intermediate Dimension: 9216
- LM Output: 248320 (Tied to token embedding)
- MTP: trained with multi-steps
Context Length: 262,144 natively and extensible up to 1,010,000 tokens.

💎 Project Context — "A Rhizome in Motion"

Downloads last month: 16

Safetensors

Model size

5B params

Tensor type

BF16

F32