Instructions to use Nerdsking/nerdsking-python-coder-3B-i with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Nerdsking/nerdsking-python-coder-3B-i with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="Nerdsking/nerdsking-python-coder-3B-i",
	filename="nerdsking-python-coder-3B-i_Q5_k_m.gguf",
)

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

Notebooks
Google Colab
Kaggle
Local Apps

llama.cpp

How to use Nerdsking/nerdsking-python-coder-3B-i with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf Nerdsking/nerdsking-python-coder-3B-i:Q5_K_M
# Run inference directly in the terminal:
llama-cli -hf Nerdsking/nerdsking-python-coder-3B-i:Q5_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf Nerdsking/nerdsking-python-coder-3B-i:Q5_K_M
# Run inference directly in the terminal:
llama-cli -hf Nerdsking/nerdsking-python-coder-3B-i:Q5_K_M

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 Nerdsking/nerdsking-python-coder-3B-i:Q5_K_M
# Run inference directly in the terminal:
./llama-cli -hf Nerdsking/nerdsking-python-coder-3B-i:Q5_K_M

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 Nerdsking/nerdsking-python-coder-3B-i:Q5_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf Nerdsking/nerdsking-python-coder-3B-i:Q5_K_M

Use Docker

docker model run hf.co/Nerdsking/nerdsking-python-coder-3B-i:Q5_K_M

LM Studio
Jan

vLLM

How to use Nerdsking/nerdsking-python-coder-3B-i with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Nerdsking/nerdsking-python-coder-3B-i"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Nerdsking/nerdsking-python-coder-3B-i",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Nerdsking/nerdsking-python-coder-3B-i:Q5_K_M

Ollama
How to use Nerdsking/nerdsking-python-coder-3B-i with Ollama:
```
ollama run hf.co/Nerdsking/nerdsking-python-coder-3B-i:Q5_K_M
```

Unsloth Studio new

How to use Nerdsking/nerdsking-python-coder-3B-i 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 Nerdsking/nerdsking-python-coder-3B-i 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 Nerdsking/nerdsking-python-coder-3B-i to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for Nerdsking/nerdsking-python-coder-3B-i to start chatting

Pi new

How to use Nerdsking/nerdsking-python-coder-3B-i with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf Nerdsking/nerdsking-python-coder-3B-i:Q5_K_M

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "Nerdsking/nerdsking-python-coder-3B-i:Q5_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use Nerdsking/nerdsking-python-coder-3B-i with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf Nerdsking/nerdsking-python-coder-3B-i:Q5_K_M

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default Nerdsking/nerdsking-python-coder-3B-i:Q5_K_M

Run Hermes

hermes

Docker Model Runner
How to use Nerdsking/nerdsking-python-coder-3B-i with Docker Model Runner:
```
docker model run hf.co/Nerdsking/nerdsking-python-coder-3B-i:Q5_K_M
```

Lemonade

How to use Nerdsking/nerdsking-python-coder-3B-i with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull Nerdsking/nerdsking-python-coder-3B-i:Q5_K_M

Run and chat with the model

lemonade run user.nerdsking-python-coder-3B-i-Q5_K_M

List all available models

lemonade list

nerdsking-python-coder-3B-i / README.md

Nerdsking

Update README.md

75114f1 verified about 1 month ago

preview code

raw

history blame contribute delete

6.5 kB

metadata

license: apache-2.0
language:
  - en
  - pt
  - zh
  - es
  - de
  - it
  - fr
metrics:
  - type:
      HumanEval zero-shot pass@1: null
    value:
      '88.41': null
base_model:
  - Qwen/Qwen2.5-Coder-3B-Instruct
pipeline_tag: text-generation
tags:
  - code

Model Details

Nerdsking-python-coder-3B-i is a 3B parameter partially uncensored model focused in coding. It was massively trained in python, therefore despite the fact it can code in other languages as well, the performance will be not in the same level as the one achieved while using python.

Key Characteristics:

Parameter count: 3B
Primary domain: Python programming
Secondary capabilities: General coding, technical English
Training focus: Python logic, standard library usage, algorithmic reasoning
Alignment: Partially uncensored (developer-oriented)

Nerdsking Python Coder Family

🧠 Nerdsking Python Coder 3B-i
🧠 Nerdsking Python Coder 7B-i

Benchmark

After intense refining, Nerdsking-python-coder-3B-i has achieved 88.41 in HumanEval (bf16), ranking it amongst the highest-performing Python-focused 3B models ever reported on HumanEval. Surpassing even much bigger models in that area.

Benchmark details (164 tasks.):

official HumanEval execution protocol: test suites executed via exec()
zero-shot pass@1
dtype == "bfloat16"
temperature = 0.1
do_sample = False
evaluated on fully merged weights
Prompting: Chat-formatted with a fixed system prompt (“You are an expert Python coding assistant.”)
Quantization: None (unquantized weights - bf16)
The configuration above is completely disclosed to support reproducibility and fair comparison.

Note: Quantized variants (INT4/INT6) may exhibit lower HumanEval scores due to reduced numerical precision.

IMPORTANT: 5 "errors" from the model during the benchmark were mere "import errors" (missing imports: reduce, Optional, List, etc ), but the logic was perfect, therefore:

the model reasoning was right
the failure is syntactic / boilerplate, not conceptual

We did not considered it for our score, but "if" considered those extra 5 questions as correct, our benchmark would be much higher.

Comparison Table with models of the same size

Model name	Approx. HumanEval Pass@1 (%)	Notes / Source
Nerdsking-python-coder-3B-i	88.41	Evaluated score (zero-shot, strict HumanEval pass@1, using unquantized weigths bf16)
StarCoder2-3B	~33.6	Reported in third-party performance overview; may differ by protocol
Qwen/Qwen3-4B-GGUF (bigger size model)	~73%	Indicative proxy from published code-task performance breakdowns (not a strict HumanEval pass@1)
Wizard Coder 3B*	~31.6 (estimate)	Indicative proxy from published code-task performance breakdowns (not a strict HumanEval pass@1)
CodeLlama 7B‑Python (much bigger size model)	~74 %	Indicative proxy from published code-task performance breakdowns (not a strict HumanEval pass@1)

*Estimated/proxy values where standardized HumanEval pass@1 was not published in those 3 models. Scores can vary with prompt format, decoding params, and harness.

TIME Comparison Table with models of Bigger size WITH "thinking mode"

Nanbeige/Nanbeige4.1-3B (4 billion model, "marketed" as 3B billions), with 92% HumanEval accuracy in Python (FP16)

Time to finish 160 questions from HumanEval: 22263.80s --- MORE THAN SIX HOURS !!! (in a RTX 5060 ti 16gb)

Nerdsking/nerdsking-python-coder-3B-i (real 3 billion model) with 88,41% HumanEval accuracy in Python (FP16)

Time to finish 160 questions from HumanEval: 210.51s --- THREE MINUTES AND HALF !!! (in a RTX 5060 ti 16gb)

Benchmark tool used

https://github.com/nerdskingcom/gguf-humaneval-benchmark

Install it using:


 pip install gguf-humaneval-benchmark

Instructions after install:


 gguf-humaneval-benchmark --help

S.o.n.n.

The model was treated under "s.o.n.n." (single omni neural network), a concept created by IPMN at Nerdsking.com that is both a precise way of fine tunning/altering existing models, as well a foundational concept for a broader AI architecture standard currently under active research and development.

When applied to pre-existing models, allows:

parameter-preserving refinement methodology
focused global behavioral shaping, instead of task-local adapters
avoidance of fragmentation, common in multi-adapter or task-siloed approaches

Quick Start (Inference)


from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "Nerdsking/Nerdsking-python-coder-3B-i"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype="bfloat16",
    device_map="auto"
)
prompt = "Write a Python function that checks if a number is prime."
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=200)

print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Ethical & Safety Notes

This model is intended for technical and/or research use. Due to relaxed alignment constraints, outputs should be reviewed before deployment in production or public-facing systems.

Citation

If you use this model in research or benchmarking, please cite:

Nerdsking-python-coder-3B-i, Iran Necho (IPMN) / Nerdsking.com