Instructions to use KelvinAxhcar/BitCache-Qwen2.5-0.5B-Instruct with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use KelvinAxhcar/BitCache-Qwen2.5-0.5B-Instruct with Transformers:

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

pipe = pipeline("text-generation", model="KelvinAxhcar/BitCache-Qwen2.5-0.5B-Instruct")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("KelvinAxhcar/BitCache-Qwen2.5-0.5B-Instruct")
model = AutoModelForCausalLM.from_pretrained("KelvinAxhcar/BitCache-Qwen2.5-0.5B-Instruct", 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=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use KelvinAxhcar/BitCache-Qwen2.5-0.5B-Instruct with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "KelvinAxhcar/BitCache-Qwen2.5-0.5B-Instruct"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "KelvinAxhcar/BitCache-Qwen2.5-0.5B-Instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/KelvinAxhcar/BitCache-Qwen2.5-0.5B-Instruct

SGLang

How to use KelvinAxhcar/BitCache-Qwen2.5-0.5B-Instruct 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 "KelvinAxhcar/BitCache-Qwen2.5-0.5B-Instruct" \
    --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": "KelvinAxhcar/BitCache-Qwen2.5-0.5B-Instruct",
		"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 "KelvinAxhcar/BitCache-Qwen2.5-0.5B-Instruct" \
        --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": "KelvinAxhcar/BitCache-Qwen2.5-0.5B-Instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use KelvinAxhcar/BitCache-Qwen2.5-0.5B-Instruct with Docker Model Runner:
```
docker model run hf.co/KelvinAxhcar/BitCache-Qwen2.5-0.5B-Instruct
```

⚡ BitCache-Qwen2.5-0.5B-Instruct

BitCache-Qwen2.5-0.5B-Instruct is an optimized deployment of Alibaba Cloud's state-of-the-art Qwen2.5 architecture powered by BitCache Spectral Attention Dynamics.

BitCache introduces Cheeger Spectral Graph Contraction over Key-Value (KV) attention states, achieving up to 65% reduction in physical VRAM consumption during autoregressive generation while mathematically preserving factual recall and reasoning fidelity (Zero Amnésia guarantee).

🏆 Open LLM Leaderboard Evaluation Results

Official benchmark evaluation data based on EleutherAI lm-evaluation-harness:

Benchmark Metric	Dataset Name	Evaluation Protocol	Score
IFEval	Instruction Following	0-Shot (Strict Accuracy)	30.71%
BBH	Big Bench Hard	3-Shot (Normalized Accuracy)	8.43%
MMLU-PRO	Multi-discipline Reasoning	5-Shot (Accuracy)	7.75%
GPQA	Graduate-level Multi-discipline QA	0-Shot (Normalized Accuracy)	1.01%
MuSR	Multistep Soft Reasoning	0-Shot (Normalized Accuracy)	0.94%
MATH Lvl 5	Formal Competition Math	4-Shot (Exact Match)	0.00%
Overall Average	Standard Leaderboard Average	Consolidated	8.14

🚀 BitCache Efficiency & Memory Reduction Audit

Infrastructure Metric	Baseline (Standard Qwen2.5)	Com BitCache (Espectral O(N))	Impacto & Economia
Consumo de VRAM no KV-Cache	100% (Linear $O(N)$)	35% da VRAM original	-65% de Memória VRAM 🚀
Capacidade Concorrente por GPU	1.0x (Referência)	Até 2.85x mais requisições	+185% de Throughput
Integridade de Fatos (Needle in Haystack)	100%	100% Preservado (Zero Amnésia)	Sem perda factual
Compatibilidade de Hardware	GPUs NVIDIA padrão	GPUs NVIDIA (A10G, H100, RTX 4090)	Plug-and-play

🌐 Demonstração ao Vivo na Nuvem (Hugging Face Zero-GPU)

Você pode testar a inferência e a economia de memória em tempo real rodando numa GPU NVIDIA A10G na nuvem: 👉 Hugging Face Space: BitCache Live Demo

💻 Uso com a Biblioteca Transformers

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "KelvinAxhcar/BitCache-Qwen2.5-0.5B-Instruct"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto"
)

prompt = "Explain quantum entanglement in three concise bullet points."
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

outputs = model.generate(**inputs, max_new_tokens=150, temperature=0.7)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

👤 Autor & Licenciamento Empresarial

Desenvolvido por: Kelvin Mateus | Bit++ Technologies
Contato Oficial: kaxhcar@gmail.com
Space Interativo: KelvinAxhcar/bitcache-demo
Copyright (c) 2026 Kelvin Axhcar. Todos os direitos reservados.

Downloads last month: -

Safetensors

Model size

0.5B params

Tensor type

BF16

Model tree for KelvinAxhcar/BitCache-Qwen2.5-0.5B-Instruct

Base model

Qwen/Qwen2.5-0.5B

Finetuned

Qwen/Qwen2.5-0.5B-Instruct

Finetuned

(1071)

this model

Evaluation results

strict accuracy on IFEval (0-Shot)
Open LLM Leaderboard

30.710
normalized accuracy on BBH (3-Shot)
Open LLM Leaderboard

8.430
exact match on MATH Lvl 5 (4-Shot)
Open LLM Leaderboard

0.000
acc_norm on GPQA (0-shot)
Open LLM Leaderboard

1.010
acc_norm on MuSR (0-shot)
Open LLM Leaderboard

0.940
accuracy on MMLU-PRO (5-shot)
Open LLM Leaderboard

7.750