Instructions to use Orivael-SRD-Lab/DeepSeek-R1-Distill-Qwen-1.5B-srd4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Orivael-SRD-Lab/DeepSeek-R1-Distill-Qwen-1.5B-srd4 with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="Orivael-SRD-Lab/DeepSeek-R1-Distill-Qwen-1.5B-srd4",
	filename="Distill-Qwen-1.5B-srd4_q4km.gguf",
)

llm.create_chat_completion(
	messages = [
		{
			"role": "user",
			"content": "What is the capital of France?"
		}
	]
)

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use Orivael-SRD-Lab/DeepSeek-R1-Distill-Qwen-1.5B-srd4 with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf Orivael-SRD-Lab/DeepSeek-R1-Distill-Qwen-1.5B-srd4
# Run inference directly in the terminal:
llama cli -hf Orivael-SRD-Lab/DeepSeek-R1-Distill-Qwen-1.5B-srd4

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf Orivael-SRD-Lab/DeepSeek-R1-Distill-Qwen-1.5B-srd4
# Run inference directly in the terminal:
llama cli -hf Orivael-SRD-Lab/DeepSeek-R1-Distill-Qwen-1.5B-srd4

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf Orivael-SRD-Lab/DeepSeek-R1-Distill-Qwen-1.5B-srd4
# Run inference directly in the terminal:
./llama-cli -hf Orivael-SRD-Lab/DeepSeek-R1-Distill-Qwen-1.5B-srd4

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf Orivael-SRD-Lab/DeepSeek-R1-Distill-Qwen-1.5B-srd4
# Run inference directly in the terminal:
./build/bin/llama-cli -hf Orivael-SRD-Lab/DeepSeek-R1-Distill-Qwen-1.5B-srd4

Use Docker

docker model run hf.co/Orivael-SRD-Lab/DeepSeek-R1-Distill-Qwen-1.5B-srd4

LM Studio
Jan

vLLM

How to use Orivael-SRD-Lab/DeepSeek-R1-Distill-Qwen-1.5B-srd4 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Orivael-SRD-Lab/DeepSeek-R1-Distill-Qwen-1.5B-srd4"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Orivael-SRD-Lab/DeepSeek-R1-Distill-Qwen-1.5B-srd4",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Orivael-SRD-Lab/DeepSeek-R1-Distill-Qwen-1.5B-srd4

Ollama
How to use Orivael-SRD-Lab/DeepSeek-R1-Distill-Qwen-1.5B-srd4 with Ollama:
```
ollama run hf.co/Orivael-SRD-Lab/DeepSeek-R1-Distill-Qwen-1.5B-srd4
```

Unsloth Studio

How to use Orivael-SRD-Lab/DeepSeek-R1-Distill-Qwen-1.5B-srd4 with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for Orivael-SRD-Lab/DeepSeek-R1-Distill-Qwen-1.5B-srd4 to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for Orivael-SRD-Lab/DeepSeek-R1-Distill-Qwen-1.5B-srd4 to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for Orivael-SRD-Lab/DeepSeek-R1-Distill-Qwen-1.5B-srd4 to start chatting

Atomic Chat new
Docker Model Runner
How to use Orivael-SRD-Lab/DeepSeek-R1-Distill-Qwen-1.5B-srd4 with Docker Model Runner:
```
docker model run hf.co/Orivael-SRD-Lab/DeepSeek-R1-Distill-Qwen-1.5B-srd4
```

Lemonade

How to use Orivael-SRD-Lab/DeepSeek-R1-Distill-Qwen-1.5B-srd4 with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull Orivael-SRD-Lab/DeepSeek-R1-Distill-Qwen-1.5B-srd4

Run and chat with the model

lemonade run user.DeepSeek-R1-Distill-Qwen-1.5B-srd4-{{QUANT_TAG}}

List all available models

lemonade list

DeepSeek-R1-Distill-1.5B · SRD4 Q4_K_M

Standard GGUF — drop into any llama.cpp build, no custom kernels.

Base model: deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
Quantization: SRD4 → Q4_K_M
File size: ~1.1 GB

What is SRD?

Standard Q4_K_M loses information systematically. Stochastic Residual Dithering (SRD) computes an INT8 residual (D8) before quantization. At load time the corrected weights are: W ≈ W4 + D8 × S8. Inference speed is identical to vanilla Q4_K_M after load.

SRD targets the reasoning layers (40–77% of depth, layers 11–21 of 28 in this model). This is a distilled reasoning model — the chain-of-thought reasoning layers are the highest-leverage SRD target. Selective correction focuses D8 on the depth range where multi-step inference coherence is encoded.

Benchmark results

Evaluated on TruthfulQA MC1 (817 questions) and WikiText-2 perplexity.

Built 2025-06-08 — benchmarks pending.

Mode	TruthfulQA MC1 ↑	Δ vs baseline	D8 overhead
Baseline Q4_K_M	TBD	—	0 MB
Selective SRD (reasoning layers)	TBD	TBD	TBD
Full SRD (all layers)	TBD	TBD	TBD

WikiText-2 PPL: TBD

Usage

Note: DeepSeek-R1 outputs <think>...</think> blocks before the final answer. These are reasoning traces, not errors — they are part of the model's chain-of-thought and can be stripped or displayed depending on your application.

llama-cli -m deepseek_r1_1b5_q4km.gguf \
  -p "<|User|>What is 15% of 840?<|Assistant|><think>" \
  --n-predict 400

How it was built

from research.quant.quantize_model import quantize_hf_model_inplace
from transformers import AutoModelForCausalLM

model = AutoModelForCausalLM.from_pretrained(
    "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B", torch_dtype=torch.float16
)
quantize_hf_model_inplace(model, alpha=1.0, group_size=64)
# → converted to GGUF Q4_K_M via llama.cpp convert_hf_to_gguf.py

Pipeline: orivael-dev/axiom — branch claude/srd-prototype-benchmark-JRtv1

Contribute results

Run llama-perplexity on WikiText-2 and open a Discussion with:

Hardware (CPU / CUDA / Metal / ROCm)
Perplexity score
Tokens/sec

Downloads last month: 122

GGUF

Model size

2B params

Architecture

qwen2

Hardware compatibility

We're not able to determine the quantization variants.

View all variants

Model tree for Orivael-SRD-Lab/DeepSeek-R1-Distill-Qwen-1.5B-srd4

Base model

deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B

Quantized

(241)

this model