---
language:
- en
license: apache-2.0
tags:
- moe
- casual-llm
- low-latency
- kitefish
- reasoning
- reasoning-l1
---

# Reasoning-L1-10B

**Reasoning-L1-10B** is a compact, efficient casual language model by KiteFishAI,
designed for low-latency inference on consumer hardware with strong reasoning
and agentic capabilities.

## Model Details

| Property | Value |
|---|---|
| Model | Reasoning-L1 |
| Parameters | ~11.4B |
| Architecture | Mixture-of-Experts (MoE) |
| VRAM required | ~8GB |
| Context length | 131K tokens |
| License | Apache 2.0 |

## Capabilities

- Configurable reasoning effort (low / medium / high)
- Full chain-of-thought support
- Function calling and structured outputs
- Web browsing and Python code execution (agentic)
- Multilingual support

## Usage

```python
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model = AutoModelForCausalLM.from_pretrained(
    "KiteFishAI/Reasoning-L1-10B",
    torch_dtype=torch.float16,
    device_map="auto",
    trust_remote_code=True,
)
tokenizer = AutoTokenizer.from_pretrained(
    "KiteFishAI/Reasoning-L1-10B",
    trust_remote_code=True,
)

inputs = tokenizer("Hello! What can you do?", return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=200)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```

## Ollama

```bash
ollama pull KiteFishAI/Reasoning-L1-10B
ollama run KiteFishAI/Reasoning-L1-10B
```

## License

Apache 2.0 — free to use, modify, and distribute commercially.

---
*Built by [KiteFishAI](https://huggingface.co/KiteFishAI)*