Image-Text-to-Text
GGUF
llama.cpp
rocm
amd
rocmfp4
rocmfpx
strix-halo
amd-strix-halo
gfx1151
ryzen-ai-max
ryzen-ai-max-395
radeon-8060s
Mixture of Experts
reasoning
multimodal
vision
nex
qwen3.5
quantized
conversational
Instructions to use kingjones777/Nex-N2.5-mini-ROCmFP4-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- llama.cpp
How to use kingjones777/Nex-N2.5-mini-ROCmFP4-GGUF 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 kingjones777/Nex-N2.5-mini-ROCmFP4-GGUF:Q4_0 # Run inference directly in the terminal: llama cli -hf kingjones777/Nex-N2.5-mini-ROCmFP4-GGUF:Q4_0
Install from WinGet (Windows)
winget install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama serve -hf kingjones777/Nex-N2.5-mini-ROCmFP4-GGUF:Q4_0 # Run inference directly in the terminal: llama cli -hf kingjones777/Nex-N2.5-mini-ROCmFP4-GGUF:Q4_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 kingjones777/Nex-N2.5-mini-ROCmFP4-GGUF:Q4_0 # Run inference directly in the terminal: ./llama-cli -hf kingjones777/Nex-N2.5-mini-ROCmFP4-GGUF:Q4_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 kingjones777/Nex-N2.5-mini-ROCmFP4-GGUF:Q4_0 # Run inference directly in the terminal: ./build/bin/llama-cli -hf kingjones777/Nex-N2.5-mini-ROCmFP4-GGUF:Q4_0
Use Docker
docker model run hf.co/kingjones777/Nex-N2.5-mini-ROCmFP4-GGUF:Q4_0
- LM Studio
- Jan
- vLLM
How to use kingjones777/Nex-N2.5-mini-ROCmFP4-GGUF with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "kingjones777/Nex-N2.5-mini-ROCmFP4-GGUF" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "kingjones777/Nex-N2.5-mini-ROCmFP4-GGUF", "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/kingjones777/Nex-N2.5-mini-ROCmFP4-GGUF:Q4_0
- Ollama
How to use kingjones777/Nex-N2.5-mini-ROCmFP4-GGUF with Ollama:
ollama run hf.co/kingjones777/Nex-N2.5-mini-ROCmFP4-GGUF:Q4_0
- Unsloth Desktop
- Pi
How to use kingjones777/Nex-N2.5-mini-ROCmFP4-GGUF with Pi:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf kingjones777/Nex-N2.5-mini-ROCmFP4-GGUF:Q4_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": "kingjones777/Nex-N2.5-mini-ROCmFP4-GGUF:Q4_0" } ] } } }Run Pi
# Start Pi in your project directory: pi
- Docker Model Runner
How to use kingjones777/Nex-N2.5-mini-ROCmFP4-GGUF with Docker Model Runner:
docker model run hf.co/kingjones777/Nex-N2.5-mini-ROCmFP4-GGUF:Q4_0
- Lemonade
How to use kingjones777/Nex-N2.5-mini-ROCmFP4-GGUF with Lemonade:
Pull the model
# Download Lemonade from https://lemonade-server.ai/ lemonade pull kingjones777/Nex-N2.5-mini-ROCmFP4-GGUF:Q4_0
Run and chat with the model
lemonade run user.Nex-N2.5-mini-ROCmFP4-GGUF-Q4_0
List all available models
lemonade list
- Hermes Agent
How to use kingjones777/Nex-N2.5-mini-ROCmFP4-GGUF with Hermes Agent:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf kingjones777/Nex-N2.5-mini-ROCmFP4-GGUF:Q4_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 kingjones777/Nex-N2.5-mini-ROCmFP4-GGUF:Q4_0
Run Hermes
hermes
- Atomic Chat
- OpenClaw
How to use kingjones777/Nex-N2.5-mini-ROCmFP4-GGUF with OpenClaw:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf kingjones777/Nex-N2.5-mini-ROCmFP4-GGUF:Q4_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 "kingjones777/Nex-N2.5-mini-ROCmFP4-GGUF:Q4_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"
File size: 9,875 Bytes
4294de3 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 | #!/usr/bin/env python3
"""Stage (hardlinks), upload, and byte-verify the two Nex-N2.5-mini repos.
usage: nex_publish.py stage|upload|verify|readme <std|imat> [README path]
Upload through upload_watchdog.sh (40G cap, stall kill + retry) - see nex_cards.sh upload
Env: default HF_HOME (token); HF_XET_CACHE / TMPDIR pinned to /mnt/models by the caller."""
import hashlib, json, os, shutil, sys, urllib.request
W = "/mnt/models/nex-n2.5-mini"; N = "Nex-N2.5-mini"; U = "kingjones777"
REPOS = {"std": f"{U}/{N}-ROCmFP4-GGUF", "imat": f"{U}/{N}-ROCmFP4-imatrix-GGUF"}
def pubname(fn):
"""Published filename: the quant token must be hyphen-delimited or the Hub cannot parse the variant
(`Q4_0_ROCMFP4_STRIX_LEAN` reads as the garbage label `Q4_0_ROCMFP`)."""
return fn.replace("Q4_0_ROCMFP4_", "Q4_0-ROCmFP4-")
MMPROJ = {f"mmproj-{N}-BF16.gguf": (f"{W}/out", f"mmproj-{N}-BF16.gguf")}
# the stock chat template plus one line that maps enable_thinking onto reasoning_effort (served with
# --chat-template-file; see the card's Reasoning controls) - shipped in both repos
TPL = {"chat_template_enable_thinking.jinja": (f"{W}/tpl", "chat_template_enable_thinking.jinja")}
TIERS = ("Q4_0_ROCMFP4_STRIX_LEAN", "Q4_0_ROCMFP4_COHERENT", "Q4_0_ROCMFP4_FAST")
FILES = {
"std": {**{pubname(f"{N}-{t}.gguf"): (f"{W}/out", f"{N}-{t}.gguf") for t in TIERS}, **MMPROJ, **TPL},
"imat": {**{pubname(f"{N}-imatrix-{t}.gguf"): (f"{W}/out-imat", f"{N}-imatrix-{t}.gguf") for t in TIERS},
f"{N}.imatrix": (f"{W}/imat", f"{N}.imatrix"), **MMPROJ, **TPL},
}
P = "recipe/pipeline/"
SCRIPTS = ("nex_download.sh", "verify_download.py", "nex_phase1.sh", "nex_phase2.sh", "diag_bf16.sh", "nex_phase2b.sh",
"nex_harness.py", "nex_bench.py", "nex_sizing.sh",
"nex_cachegate.py", "readback.py", "nex_aggregate.py", "nex_judge.py", "nex_render.py", "nex_publish.py",
"nex_cards.sh", "upload_watchdog.sh", "nex_seats.sh",
# tool-call / reasoning diagnosis (stock template 6/14) and the template fix
"diag_tools_run.sh", "nex_tools_diag.py", "nex_reasoning_probe.py", "nex_tools_tpl.py",
"nex_seat_default_probe.py", "run_tools_c1.sh", "run_tools_roff.sh", "nex_refresh_cards.sh", "nex_publish_finish.sh")
LOGS = (
"D2_verify_download.log", "phase1.log", "phase2.log", "Q_sizes.log", "N6_bench.log", "N7_sizing.log",
"b_n-vision-q106-faon.log", "b_n-vision-q106-faoff.log",
"C1_convert.log", "C2_mmproj.log", "C_readback.log",
"Q1_q106.log", "Q1_q102.log", "Q1_q103.log", "Q_readback.log",
"N1c_ppl_bf16_cpu.log", "N2c_imatrix_cpu.log",
# the stopped first attempt on ROCm0 and the backend diagnosis that followed (the Known issues evidence)
"N1_ppl_bf16.log", "diag_ppl_q106_rocm_c4.log", "diag_bf16.log", "diag_bf16_rocm_faoff.log", "diag_bf16_vk_faon.log",
"diag_q106_vk_faon.log", "diag_bf16_cpu.log", "diag_bf16_purecpu_c1.log",
"N3_q106i.log", "N3_q102i.log", "N3_q103i.log", "N3_readback.log",
*(f"{p}_kld_{t}.log" for p in ("N4", "N4v") for t in ("q106", "q102", "q103", "q106i", "q102i", "q103i")),
"N5_kld_q106_repeat.log", "N5v_kld_q106_repeat.log", "b_n-c3-q106.log",
"b_n-tools-q106.log", "diag_tools.log", "diag_tools_server.log",
"probe_reasoning.log", *(f"probe_reasoning_{c}.log" for c in
("default", "fmt-deepseek", "srv-kwargs-high", "reasoning-on", "tpl-enable-thinking")),
"N6t_tools_tpl.log", "b_n-tools-q106-tpl.log", "b_n-tools-q106-tpl-probe.log",
"N6t_tools_tpl_medium.log", "b_n-tools-q106-tpl-medium.log", "b_n-tools-q106-tpl-medium-probe.log",
"probe_seat_default.log", "probe_seat_default_C1.log", "probe_seat_default_C2.log",
"N6t_tools_c1.log", "b_n-tools-q106-c1.log", "b_n-tools-q106-c1-probe.log", "b_n-vision-q106-c1-faon.log",
"N6t_tools_roff.log", "b_n-tools-q106-roff.log", "b_n-tools-q106-roff-r2.log", "b_n-tools-q106-roff-r3.log",
"b_n-tools-q106-roff-probe.log", "b_n-vision-q106-roff-faon.log",
"N8_unice.log", "N8a_seats.log", "N8b_seats.log", "N8c_seats.log", "N8d_seats.log",
)
RAW = ("nex_repeat.jsonl", "nex_reference.jsonl", "nex_bench.jsonl", "nex_sizing.jsonl", "hub_check.json",
"nex_tools_diag.json", "nex_reasoning_probe.json", "nex_seat_default_probe.json", "nex_template_shim.json",
"nex_template_shim_medium.json", "nex_template_shim_c1.json", "nex_template_shim_roff.json",
"nex_seats.jsonl", "nex_seats_plan.json")
TEMPLATES = ("chat_template_enable_thinking_medium.jinja", "chat_template_enable_thinking_v2.jinja") # tested, not used
RECIPE = {
"recipe/results_summary.json": "results/summary.json",
**{P + x: x for x in SCRIPTS},
**{f"recipe/raw/{x}": f"results/{x}" for x in RAW},
**{f"recipe/logs/{x}": f"logs/{x}" for x in LOGS},
**{f"recipe/templates/{x}": f"tpl/{x}" for x in TEMPLATES},
"recipe/templates/chat_template_stock.jinja": "hf/chat_template.jinja",
}
def separation_errors(kind, names):
"""The imatrix build is its OWN repo (King): no imatrix weights in the standard repo, no standard weights in the
imatrix repo. The vision projector is the only model file both carry."""
errs = []
for fn in names:
if not fn.endswith(".gguf") or fn.startswith("mmproj-"):
continue
is_imat = "-imatrix-" in fn
if kind == "std" and is_imat:
errs.append(f"imatrix model file in the standard repo: {fn}")
if kind == "imat" and not is_imat:
errs.append(f"standard model file in the imatrix repo: {fn}")
if kind == "imat" and f"{N}.imatrix" not in names:
errs.append("imatrix repo is missing the .imatrix file")
return errs
assert REPOS["std"] != REPOS["imat"]
assert set(FILES["std"]) & set(FILES["imat"]) == set(MMPROJ) | set(TPL), "only the projector + template may be shared"
for _k in FILES:
assert not separation_errors(_k, list(FILES[_k])), separation_errors(_k, list(FILES[_k]))
def git_blob_sha1(p):
"""What the Hub reports as `oid` for a file stored in plain git (no LFS/Xet pointer)."""
h = hashlib.sha1(b"blob %d\0" % os.path.getsize(p))
with open(p, "rb") as fh:
for b in iter(lambda: fh.read(64 << 20), b""):
h.update(b)
return h.hexdigest()
def sha(p):
h = hashlib.sha256()
with open(p, "rb") as fh:
for b in iter(lambda: fh.read(64 << 20), b""):
h.update(b)
return h.hexdigest()
kind = sys.argv[2]; stage = f"{W}/hf-upload/{kind}"; repo = REPOS[kind]
if sys.argv[1] == "stage":
missing = [f"{d}/{src}" for fn, (d, src) in FILES[kind].items() if not os.path.exists(f"{d}/{src}")] + \
[src for src in RECIPE.values() if not os.path.exists(f"{W}/{src}")]
if missing:
raise SystemExit(f"STAGE ABORT - missing: {missing}")
shutil.rmtree(stage, ignore_errors=True)
for fn, (d, src) in FILES[kind].items():
os.makedirs(stage, exist_ok=True); os.link(f"{d}/{src}", f"{stage}/{fn}")
for dst, src in RECIPE.items():
os.makedirs(os.path.dirname(f"{stage}/{dst}"), exist_ok=True); shutil.copy2(f"{W}/{src}", f"{stage}/{dst}")
sums = {fn: sha(f"{stage}/{fn}") for fn in sorted(FILES[kind])}
with open(f"{stage}/SHA256SUMS", "w") as fh:
for fn, h in sums.items():
fh.write(f"{h} {fn}\n")
sizes = {x: os.path.getsize(f"{stage}/{x}") for x in sums}
json.dump(sizes, open(f"{stage}/../staged_sizes_{kind}.json", "w"))
print(json.dumps({"stage": stage, "files": len(sums), "bytes": sum(sizes.values()), "recipe_files": len(RECIPE)}))
elif sys.argv[1] == "upload":
from huggingface_hub import HfApi
api = HfApi()
print("whoami:", api.whoami()["name"], flush=True)
api.create_repo(repo, repo_type="model", private=False, exist_ok=True)
api.upload_large_folder(repo_id=repo, repo_type="model", folder_path=stage,
ignore_patterns=["README.md", ".cache/**"], num_workers=2)
print("UPLOAD_DONE", repo, flush=True)
elif sys.argv[1] == "verify":
tree = json.load(urllib.request.urlopen(f"https://huggingface.co/api/models/{repo}/tree/main?recursive=true", timeout=60))
remote = {t["path"]: t for t in tree if t.get("type") == "file"}
local = dict(l.split()[::-1] for l in open(f"{stage}/SHA256SUMS").read().splitlines())
bad = []
for fn, h in local.items():
r = remote.get(fn)
if not r:
bad.append((fn, "MISSING remote")); continue
if r["size"] != os.path.getsize(f"{stage}/{fn}"):
bad.append((fn, f"size {r['size']}")); continue
lfs = r.get("lfs")
if lfs:
if lfs.get("oid") != h:
bad.append((fn, f"sha {lfs.get('oid')} != {h[:12]}"))
elif r.get("oid") != git_blob_sha1(f"{stage}/{fn}"):
bad.append((fn, f"git oid {r.get('oid')} does not match the staged bytes"))
for fn in list(RECIPE) + ["SHA256SUMS"]:
if fn not in remote:
bad.append((fn, "MISSING remote"))
stale = sorted(x for x in remote if x not in local and x not in RECIPE and x not in ("SHA256SUMS", "README.md", ".gitattributes"))
bad += [(x, "SEPARATION") for x in separation_errors(kind, list(remote))]
print(json.dumps({"repo": repo, "checked": len(local), "bad": bad, "unexpected_remote_files": stale,
"result": "PASS" if not bad and not stale else "FAIL"}))
sys.exit(0 if not bad and not stale else 1)
elif sys.argv[1] == "readme":
from huggingface_hub import HfApi
HfApi().upload_file(path_or_fileobj=sys.argv[3], path_in_repo="README.md", repo_id=repo, repo_type="model",
commit_message="Model card: measured results, reproduction, known issues")
print("README_DONE", repo)
else:
raise SystemExit(f"usage: {sys.argv[0]} stage|upload|verify|readme <std|imat> [README path]")
|