Text Generation
Transformers
Safetensors
qwen3_5
image-text-to-text
veriloop
auto-round
community-quantization
experimental
conversational
4-bit precision
Instructions to use bowmanslayer/VeriLoop-E2-W4A16 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use bowmanslayer/VeriLoop-E2-W4A16 with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="bowmanslayer/VeriLoop-E2-W4A16") 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)# pip install -U transformers accelerate # Load model directly from transformers import AutoProcessor, AutoModelForMultimodalLM processor = AutoProcessor.from_pretrained("bowmanslayer/VeriLoop-E2-W4A16") model = AutoModelForMultimodalLM.from_pretrained("bowmanslayer/VeriLoop-E2-W4A16", 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=256) print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:])) - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use bowmanslayer/VeriLoop-E2-W4A16 with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "bowmanslayer/VeriLoop-E2-W4A16" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "bowmanslayer/VeriLoop-E2-W4A16", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/bowmanslayer/VeriLoop-E2-W4A16
- SGLang
How to use bowmanslayer/VeriLoop-E2-W4A16 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 "bowmanslayer/VeriLoop-E2-W4A16" \ --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": "bowmanslayer/VeriLoop-E2-W4A16", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'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 "bowmanslayer/VeriLoop-E2-W4A16" \ --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": "bowmanslayer/VeriLoop-E2-W4A16", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }' - Docker Model Runner
How to use bowmanslayer/VeriLoop-E2-W4A16 with Docker Model Runner:
docker model run hf.co/bowmanslayer/VeriLoop-E2-W4A16
File size: 4,189 Bytes
711ed9f | 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 | """Frozen legacy rubric with engine field fixes, frozen questions and isolated code execution."""
import pathlib,json,subprocess,uuid,os
import legacy_capability as h
from protocol import normalize_content
ROOT=pathlib.Path(__file__).parent
load_legacy=h.load_items
def load_items(task,n,seed):
data=json.loads((ROOT/'questions.json').read_text())
assert seed==data['seed']
items=data['tasks'][task]
assert len(items)==n,(task,len(items),n)
return items
h.load_items=load_items
# Expose vLLM's current reasoning field and finish reason without changing scoring.
old_call=h.call
def call(session,args,item,spec):
class Adapter:
def post(self,*a,**kw):
r=session.post(*a,**kw)
if r.ok:
d=r.json();c=d['choices'][0];m=c['message']
raw=m.get('content') or ''
content,reasoning,mode=normalize_content(raw,c.get('finish_reason')) if args.thinking else (raw,'','thinking-disabled')
m['content']=content
m['reasoning_content']=reasoning
d.setdefault('usage',{})['content_mode']=mode
d.setdefault('usage',{})['finish_reason']=c.get('finish_reason')
r.json=lambda:d
return r
return old_call(Adapter(),args,item,spec)
h.call=call
def score_code(text,gold,timeout=15):
code=h._extract_code(text)
if not code.strip():return False
program='import math,re,sys,itertools,collections,heapq,functools,string\n'+code+'\n'+gold['test']+'\ncheck('+gold['entry_point']+')\n'
name='veriloop-eval-'+uuid.uuid4().hex
cmd=['docker','run','--rm','-i','--name',name,'--network','none','--read-only','--cap-drop','ALL','--security-opt','no-new-privileges','--pids-limit','64','--memory','256m','--cpus','1','--user','65534:65534','--entrypoint','python3','flashnext-base:locked-20261008','-I','-']
try:return subprocess.run(cmd,input=program,text=True,stdout=subprocess.DEVNULL,stderr=subprocess.DEVNULL,timeout=timeout).returncode==0
except subprocess.TimeoutExpired:return False
finally:subprocess.run(['docker','rm','-f',name],stdout=subprocess.DEVNULL,stderr=subprocess.DEVNULL)
h.score_code=score_code
def main():
import argparse,threading,time
from concurrent.futures import ThreadPoolExecutor
p=argparse.ArgumentParser()
for flag in ['base-url','model','label','outdir']:p.add_argument('--'+flag,required=True)
p.add_argument('--tasks',default='all');p.add_argument('--seed',type=int,default=20260816)
p.add_argument('--max-tokens',dest='max_tokens_override',type=int,default=32768)
p.add_argument('--concurrency',type=int,default=16);p.add_argument('--timeout',type=int,default=3600)
p.add_argument('--no-thinking',dest='thinking',action='store_false')
args=p.parse_args();args.n_override=0;args.temperature=0.;args.top_p=1.;args.retries=3
tasks=h.ALL_TASKS if args.tasks=='all' else args.tasks.split(',')
limit=threading.BoundedSemaphore(args.concurrency)
underlying=h.call
def bounded(*a,**kw):
with limit:return underlying(*a,**kw)
h.call=bounded
out=pathlib.Path(args.outdir);out.mkdir(exist_ok=True)
summary={'label':args.label,'model':args.model,'base_url':args.base_url,'seed':args.seed,'thinking':args.thinking,'started':time.strftime('%F %T'),'tasks':{},'scheduler':'task workers share one request semaphore'}
def task(t):
ckpt=out/(args.label+'-'+t+'.json')
if ckpt.exists():r=json.loads(ckpt.read_text())
else:
print('START',t,flush=True);r=h.run_task(args,t)
if r['errors']:raise RuntimeError(t+' request errors')
ckpt.write_text(json.dumps(r,ensure_ascii=False,indent=2))
print('RESULT',t,r['hits'],r['n'],'empty',r.get('empty'),'length',r.get('length_limited'),flush=True)
return t,{k:v for k,v in r.items() if k!='details'}
with ThreadPoolExecutor(max_workers=len(tasks)) as pool:
for t,r in pool.map(task,tasks):summary['tasks'][t]=r
summary['finished']=time.strftime('%F %T')
(out/('SUMMARY-'+args.label+'.json')).write_text(json.dumps(summary,indent=2))
if __name__=='__main__':main()
|