Image-Text-to-Text
Transformers
Safetensors
gemma4_unified
gemma4
nvfp4
fp8
compressed-tensors
vllm
speculative-decoding
experimental
conversational
8-bit precision
Instructions to use DevelopingDad/Gemma4-12B-it-Fast with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use DevelopingDad/Gemma4-12B-it-Fast with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("image-text-to-text", model="DevelopingDad/Gemma4-12B-it-Fast") 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("DevelopingDad/Gemma4-12B-it-Fast") model = AutoModelForMultimodalLM.from_pretrained("DevelopingDad/Gemma4-12B-it-Fast", 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 DevelopingDad/Gemma4-12B-it-Fast with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "DevelopingDad/Gemma4-12B-it-Fast" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "DevelopingDad/Gemma4-12B-it-Fast", "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/DevelopingDad/Gemma4-12B-it-Fast
- SGLang
How to use DevelopingDad/Gemma4-12B-it-Fast 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 "DevelopingDad/Gemma4-12B-it-Fast" \ --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": "DevelopingDad/Gemma4-12B-it-Fast", "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 "DevelopingDad/Gemma4-12B-it-Fast" \ --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": "DevelopingDad/Gemma4-12B-it-Fast", "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" } } ] } ] }' - Docker Model Runner
How to use DevelopingDad/Gemma4-12B-it-Fast with Docker Model Runner:
docker model run hf.co/DevelopingDad/Gemma4-12B-it-Fast
Download benchmarks/evaluate.py from DevelopingDad/Gemma4-12B-it-Fast: direct link, hf CLI and curl.
- Browser
- Download file 7.25 kB
-
https://huggingface.co/DevelopingDad/Gemma4-12B-it-Fast/resolve/main/benchmarks/evaluate.py
- Command line
-
hf download hf://DevelopingDad/Gemma4-12B-it-Fast/benchmarks/evaluate.py
-
curl -L -o evaluate.py https://huggingface.co/DevelopingDad/Gemma4-12B-it-Fast/resolve/main/benchmarks/evaluate.py
7.25 kB
| #!/usr/bin/env python3 | |
| # SPDX-License-Identifier: Apache-2.0 | |
| # Copyright 2026 DevelopingDad | |
| """Self-contained native API diagnostics; no external datasets or dependencies.""" | |
| import argparse | |
| import json | |
| import time | |
| import urllib.request | |
| MODEL = 'DevelopingDad/Gemma4-12B-it-Fast' | |
| def post(base, payload): | |
| request = urllib.request.Request(base.rstrip('/') + '/chat/completions', | |
| data=json.dumps(payload).encode(), headers={'Content-Type':'application/json'}) | |
| return urllib.request.urlopen(request, timeout=180) | |
| def request(messages, **kwargs): | |
| result = {'model': MODEL, 'messages': messages, 'temperature':0, 'seed':91027, | |
| 'max_tokens':256, 'chat_template_kwargs':{'enable_thinking':False}} | |
| result.update(kwargs) | |
| return result | |
| def evaluate(base): | |
| results = {'schema':'gemma4-fast-generic-api-screen/v1', | |
| 'hardware':'one NVIDIA DGX Spark GB10', 'runtime':'vLLM 0.31.0, weight-only target, MTP8, FP8 KV', | |
| 'scope':'bounded native API diagnostic; not a standardized quality benchmark', | |
| 'sampling':{'temperature':0,'seed':91027}, 'checks':[], 'generation':[]} | |
| prompts = [ | |
| ('explanation', 'Explain how condensation forms on a cold glass. Include a concrete example and explain the role of temperature.'), | |
| ('comparison', 'Compare a binary search tree with a hash table. Explain lookup, insertion, ordering and a suitable use case for each.'), | |
| ('code', 'Implement a Python function that merges two sorted integer lists. Include type hints, a docstring, and explain the time and space complexity.'), | |
| ] | |
| for name, prompt in prompts: | |
| body = request([{'role':'user','content':prompt}], max_tokens=512, stream=True, | |
| stream_options={'include_usage':True}, cache_salt='generic-public-screen-'+str(time.time_ns())) | |
| start = time.monotonic(); first = None; usage = {}; finish = None; chunks = []; reasoning = [] | |
| with post(base, body) as response: | |
| for line in response: | |
| if not line.startswith(b'data: '): continue | |
| raw = line[6:].strip() | |
| if raw == b'[DONE]': break | |
| item = json.loads(raw) | |
| if item.get('usage'): usage = item['usage'] | |
| for choice in item.get('choices', []): | |
| delta = choice.get('delta', {}) | |
| if delta.get('content'): | |
| if first is None: first = time.monotonic() | |
| chunks.append(delta['content']) | |
| if delta.get('reasoning_content'): reasoning.append(delta['reasoning_content']) | |
| if choice.get('finish_reason'): finish = choice['finish_reason'] | |
| end = time.monotonic(); tokens = usage.get('completion_tokens',0) | |
| # Completion usage can include final control tokens; this is an explicitly | |
| # approximate decode rate, not the old conservative delivery metric. | |
| results['generation'].append({'case':name,'max_tokens':512,'finish_reason':finish, | |
| 'prompt_tokens':usage.get('prompt_tokens'), 'completion_tokens':tokens, | |
| 'first_content_s':None if first is None else round(first-start,4), | |
| 'elapsed_s':round(end-start,4), | |
| 'approximate_completion_tokens_per_s_after_first_content':None if first is None else round(tokens/max(end-first,1e-9),3), | |
| 'nonempty_content':bool(''.join(chunks).strip()),'unexpected_reasoning':bool(reasoning)}) | |
| # Explicit reasoning and a completion allowance exceeding its reasoning cap. | |
| body = request([{'role':'user','content':'Calculate 137 * 24 - 89. Give the final integer.'}], | |
| max_tokens=4096, thinking_token_budget=2048, | |
| chat_template_kwargs={'enable_thinking':True}) | |
| start = time.monotonic() | |
| with post(base,body) as response: item=json.load(response) | |
| answer=item['choices'][0]['message'].get('content') or '' | |
| results['checks'].append({'case':'reasoning_arithmetic','passed':'3199' in answer.replace(',',''), | |
| 'finish_reason':item['choices'][0]['finish_reason'],'elapsed_s':round(time.monotonic()-start,4), | |
| 'completion_tokens':item.get('usage',{}).get('completion_tokens'), | |
| 'reasoning_tokens':(item.get('usage',{}).get('completion_tokens_details') or {}).get('reasoning_tokens')}) | |
| schema={'type':'object','properties':{'color':{'type':'string','enum':['blue']},'count':{'type':'integer','enum':[3]}},'required':['color','count'],'additionalProperties':False} | |
| body=request([{'role':'user','content':'Return color blue and count 3 using the supplied JSON schema.'}], | |
| response_format={'type':'json_schema','json_schema':{'name':'sample','strict':True,'schema':schema}}) | |
| with post(base,body) as response: item=json.load(response) | |
| try: passed=json.loads(item['choices'][0]['message']['content'])=={'color':'blue','count':3} | |
| except (ValueError,TypeError,KeyError): passed=False | |
| results['checks'].append({'case':'strict_json_schema','passed':passed,'finish_reason':item['choices'][0]['finish_reason']}) | |
| tool={'type':'function','function':{'name':'get_temperature','description':'Read a temperature for a named city.', | |
| 'parameters':{'type':'object','properties':{'city':{'type':'string'}},'required':['city'],'additionalProperties':False}}} | |
| messages=[{'role':'user','content':'Use the temperature tool for Paris.'}] | |
| body=request(messages,tools=[tool],tool_choice={'type':'function','function':{'name':'get_temperature'}}) | |
| with post(base,body) as response: item=json.load(response) | |
| message=item['choices'][0]['message']; calls=message.get('tool_calls') or [] | |
| passed=bool(calls) and calls[0]['function']['name']=='get_temperature' and json.loads(calls[0]['function']['arguments']).get('city','').lower()=='paris' | |
| results['checks'].append({'case':'tool_call','passed':passed,'finish_reason':item['choices'][0]['finish_reason']}) | |
| if passed: | |
| messages.extend([{'role':'assistant','content':message.get('content'),'tool_calls':calls}, | |
| {'role':'tool','tool_call_id':calls[0]['id'],'content':'{"city":"Paris","temperature_celsius":17}'}]) | |
| with post(base,request(messages,tools=[tool],tool_choice='none')) as response: item=json.load(response) | |
| answer=item['choices'][0]['message'].get('content') or '' | |
| results['checks'].append({'case':'tool_result_continuation','passed':'17' in answer,'finish_reason':item['choices'][0]['finish_reason']}) | |
| results['all_checks_passed']=all(x['passed'] for x in results['checks']) and len(results['checks'])==4 | |
| results['generation_nonempty']=all(x['nonempty_content'] for x in results['generation']) | |
| return results | |
| def main(): | |
| parser=argparse.ArgumentParser(description=__doc__) | |
| parser.add_argument('--base-url',default='http://127.0.0.1:8000/v1') | |
| parser.add_argument('--output',default='generic-results.json') | |
| args=parser.parse_args() | |
| results=evaluate(args.base_url) | |
| with open(args.output,'w') as handle: json.dump(results,handle,indent=2); handle.write('\n') | |
| print(json.dumps(results,indent=2)) | |
| return 0 if results['all_checks_passed'] and results['generation_nonempty'] else 1 | |
| if __name__=='__main__': raise SystemExit(main()) | |