Instructions to use vtava/Qwen3.5-0.8B-PDelta3-CLVR-Local32 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use vtava/Qwen3.5-0.8B-PDelta3-CLVR-Local32 with Transformers:

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

pipe = pipeline("text-generation", model="vtava/Qwen3.5-0.8B-PDelta3-CLVR-Local32")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("vtava/Qwen3.5-0.8B-PDelta3-CLVR-Local32")
model = AutoModelForCausalLM.from_pretrained("vtava/Qwen3.5-0.8B-PDelta3-CLVR-Local32", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.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(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use vtava/Qwen3.5-0.8B-PDelta3-CLVR-Local32 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "vtava/Qwen3.5-0.8B-PDelta3-CLVR-Local32"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "vtava/Qwen3.5-0.8B-PDelta3-CLVR-Local32",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/vtava/Qwen3.5-0.8B-PDelta3-CLVR-Local32

SGLang

How to use vtava/Qwen3.5-0.8B-PDelta3-CLVR-Local32 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 "vtava/Qwen3.5-0.8B-PDelta3-CLVR-Local32" \
    --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": "vtava/Qwen3.5-0.8B-PDelta3-CLVR-Local32",
		"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 "vtava/Qwen3.5-0.8B-PDelta3-CLVR-Local32" \
        --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": "vtava/Qwen3.5-0.8B-PDelta3-CLVR-Local32",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use vtava/Qwen3.5-0.8B-PDelta3-CLVR-Local32 with Docker Model Runner:
```
docker model run hf.co/vtava/Qwen3.5-0.8B-PDelta3-CLVR-Local32
```

Qwen3.5-0.8B-PDelta3-CLVR-Local32 / load_model.py

vtava

Upload verified PDelta3-CLVR checkpoint for layers [3, 7, 11]

a38f163 verified 16 days ago

Raw History Blame Contribute Delete

1.68 kB

	from __future__ import annotations

	import json
	import sys
	from pathlib import Path
	import torch
	from safetensors.torch import load_file
	from transformers import AutoTokenizer, Qwen3_5ForCausalLM

	def load_model(path=None, device="cpu", dtype=None):
	root = Path(path or Path(__file__).resolve().parent)
	for p in (root / "src", root, root / "scripts"):
	if str(p) not in sys.path:
	sys.path.insert(0, str(p))
	from train_qwen35_pdelta3_clvr_sequential import QwenPDelta3CLVRConfig, replace_full_attention_layers
	meta = json.loads((root / "tinycenn_qwen35.json").read_text())
	accepted = [int(x) for x in meta["accepted_full_attention_layers"]]
	cfg = QwenPDelta3CLVRConfig.from_dict(meta["replacement_config"])
	if dtype is None:
	dtype = torch.bfloat16 if device.startswith("cuda") and torch.cuda.is_bf16_supported() else (torch.float16 if device.startswith("cuda") else torch.float32)
	model = Qwen3_5ForCausalLM.from_pretrained(root, dtype=dtype, local_files_only=True, attn_implementation="eager")
	replace_full_attention_layers(model, cfg, accepted)
	single = root / "model.safetensors"
	if single.exists():
	model.load_state_dict(load_file(str(single), device="cpu"), strict=False)
	else:
	index = json.loads((root / "model.safetensors.index.json").read_text())
	for shard in sorted(set(index["weight_map"].values())):
	model.load_state_dict(load_file(str(root / shard), device="cpu"), strict=False)
	model.config.use_cache = False
	model.to(device).eval()
	tokenizer = AutoTokenizer.from_pretrained(root, local_files_only=True, use_fast=True)
	return model, tokenizer