README.md

---
license: apache-2.0
base_model:
- deepseek-ai/DeepSeek-R1-Distill-Qwen-14B
base_model_relation: quantized
pipeline_tag: text2text-generation
language:
- zho
- eng
- fra
- spa
- por
- deu
- ita
- rus
- jpn
- kor
- vie
- tha
- ara
---

# Elastic model: DeepSeek-R1-Distill-Qwen-14B. Fastest and most flexible models for self-serving.

Elastic models are the models produced by TheStage AI ANNA: Automated Neural Networks Accelerator. ANNA allows you to control model size, latency and quality with a simple slider movement. For each model, ANNA produces a series of optimized models:

* __XL__: Mathematically equivalent neural network, optimized with our DNN compiler.

* __L__: Near lossless model, with less than 1% degradation obtained on corresponding benchmarks.

* __M__: Faster model, with accuracy degradation less than 1.5%.

* __S__: The fastest model, with accuracy degradation less than 2%.


__Goals of elastic models:__

* Provide flexibility in cost vs quality selection for inference
* Provide clear quality and latency benchmarks
* Provide interface of HF libraries: transformers and diffusers with a single line of code
* Provide models supported on a wide range of hardware, which are pre-compiled and require no JIT.
* Provide the best models and service for self-hosting.

> It's important to note that specific quality degradation can vary from model to model. For instance, with an S model, you can have 0.5% degradation as well.

![Performance Graph](images/performance_graph.png)
-----

## Inference

To infer our models, you just need to replace `transformers` import with `elastic_models.transformers`:

```python
import torch
from transformers import AutoTokenizer
from elastic_models.transformers import AutoModelForCausalLM

# Currently we require to have your HF token
# as we use original weights for part of layers and
# model confugaration as well
model_name = "deepseek-ai/DeepSeek-R1-Distill-Qwen-14B"
hf_token = ''
device = torch.device("cuda")

# Create mode
tokenizer = AutoTokenizer.from_pretrained(
    model_name, token=hf_token
)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    token=hf_token,
    torch_dtype=torch.bfloat16,
    attn_implementation="sdpa",
    mode='S'
).to(device)
model.generation_config.pad_token_id = tokenizer.eos_token_id

# Inference simple as transformers library
prompt = "Describe basics of DNNs quantization."
messages = [
  {
    "role": "system",
    "content": "You are a search bot, answer on user text queries."
  },
  {
    "role": "user",
    "content": prompt
  }
]

chat_prompt = tokenizer.apply_chat_template(
    messages, add_generation_prompt=True, tokenize=False
)

inputs = tokenizer(chat_prompt, return_tensors="pt")
inputs.to(device)

with torch.inference_mode():
    generate_ids = model.generate(**inputs, max_length=500)

input_len = inputs['input_ids'].shape[1]
generate_ids = generate_ids[:, input_len:]
output = tokenizer.batch_decode(
    generate_ids,
    skip_special_tokens=True,
    clean_up_tokenization_spaces=False
)[0]

# Validate answer
print(f"# Q:\n{prompt}\n")
print(f"# A:\n{output}\n")
```

__System requirements:__
* GPUs: H100, L40s
* CPU: AMD, Intel
* Python: 3.10-3.12


To work with our models just run these lines in your terminal:

```shell
pip install thestage
pip install elastic_models[nvidia]\
 --index-url https://thestage.jfrog.io/artifactory/api/pypi/pypi-thestage-ai-production/simple\
 --extra-index-url https://pypi.nvidia.com\
 --extra-index-url https://pypi.org/simple

pip install flash_attn==2.7.3 --no-build-isolation
pip uninstall apex
```

Then go to [app.thestage.ai](https://app.thestage.ai), login and generate API token from your profile page. Set up API token as follows:

```shell
thestage config set --api-token <YOUR_API_TOKEN>
```

Congrats, now you can use accelerated models!

----

## Benchmarks

Benchmarking is one of the most important procedures during model acceleration. We aim to provide clear performance metrics for models using our algorithms. The `W8A8, int8 column` indicates that we applied W8A8 quantization with int8 data type to all linear layers and used the same calibration data as for ANNA. The S model achieves practically identical speed but much higher quality, as ANNA knows how to improve quantization quality on sensitive layers!

### Quality benchmarks

| Metric/Model  | S | M | L | XL | Original | W8A8, int8 |
|---------------|---|---|---|----|----------|------------|
| arc_challenge | 49.20 | 51.00 | 50.60 | 50.90 | 50.90 | 35.30 | - |
| mmlu | 73.70 | 74.30 | 74.60 | 74.80 | 74.80 | 51.50 | - |
| piqa | 77.20 | 77.90 | 78.20 | 78.60 | 78.60 | 69.70 | - |
| winogrande | 69.30 | 70.90 | 72.10 | 72.30 | 72.30 | 61.30 | - |



* **MMLU**: Evaluates general knowledge across 57 subjects including science, humanities, engineering, and more. Shows model's ability to handle diverse academic topics.
* **PIQA**: Evaluates physical commonsense reasoning through questions about everyday physical interactions. Shows model's understanding of real-world physics concepts.
* **Arc Challenge**: Evaluates grade-school level multiple-choice questions requiring reasoning. Shows model's ability to solve complex reasoning tasks.
* **Winogrande**: Evaluates commonsense reasoning through sentence completion tasks. Shows model's capability to understand context and resolve ambiguity.

### Latency benchmarks

__100 input/300 output; tok/s:__

| GPU/Model | S   | M | L | XL | Original | W8A8, int8 |
|-----------|-----|---|---|----|----------|------------|
| H100 | 118 | 105 | 95 | 77 | 39 | 123 | - |
| L40S | 40 | 35 | 31 | 24 | 22 | 41 | - |



## Links

* __Platform__: [app.thestage.ai](https://app.thestage.ai/models)
* __Subscribe for updates__: [TheStageAI X](https://x.com/TheStageAI)
<!-- * __Elastic models Github__: [app.thestage.ai](app.thestage.ai) -->
* __Contact email__: contact@thestage.ai