---
library_name: transformers
license: apache-2.0
tags:
- llm-compressor
- tiny-model
- mtp
- pr-3225
- test-fixture
---

# Inkling-0.92B-MTP

Small BF16 test model using [thinkingmachines/Inkling](https://huggingface.co/thinkingmachines/Inkling), with a toy-trained backbone and synthetic MTP weights.

## Load the model

This fixture has no published quantized variant.

```python
import torch
from transformers import InklingForConditionalGeneration

MODEL_ID = "inference-optimization/Inkling-0.92B-MTP"
model = InklingForConditionalGeneration.from_pretrained(
    MODEL_ID, dtype=torch.bfloat16, attn_implementation="eager",
)
```