Instructions to use Trivnpk/Qwen-Compiler-Mind-4B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use Trivnpk/Qwen-Compiler-Mind-4B 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 Trivnpk/Qwen-Compiler-Mind-4B:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf Trivnpk/Qwen-Compiler-Mind-4B:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf Trivnpk/Qwen-Compiler-Mind-4B:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf Trivnpk/Qwen-Compiler-Mind-4B:Q4_K_M

Build from source code

git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-app
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama serve -hf Trivnpk/Qwen-Compiler-Mind-4B:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama cli -hf Trivnpk/Qwen-Compiler-Mind-4B:Q4_K_M

Use Docker

docker model run hf.co/Trivnpk/Qwen-Compiler-Mind-4B:Q4_K_M

Llama new
LM Studio
Jan

vLLM

How to use Trivnpk/Qwen-Compiler-Mind-4B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Trivnpk/Qwen-Compiler-Mind-4B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Trivnpk/Qwen-Compiler-Mind-4B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Trivnpk/Qwen-Compiler-Mind-4B:Q4_K_M

Ollama
How to use Trivnpk/Qwen-Compiler-Mind-4B with Ollama:
```
ollama run hf.co/Trivnpk/Qwen-Compiler-Mind-4B:Q4_K_M
```
Unsloth Desktop

How to use Trivnpk/Qwen-Compiler-Mind-4B with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Trivnpk/Qwen-Compiler-Mind-4B:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "Trivnpk/Qwen-Compiler-Mind-4B:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use Trivnpk/Qwen-Compiler-Mind-4B with Docker Model Runner:
```
docker model run hf.co/Trivnpk/Qwen-Compiler-Mind-4B:Q4_K_M
```

Lemonade

How to use Trivnpk/Qwen-Compiler-Mind-4B with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull Trivnpk/Qwen-Compiler-Mind-4B:Q4_K_M

Run and chat with the model

lemonade run user.Qwen-Compiler-Mind-4B-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use Trivnpk/Qwen-Compiler-Mind-4B with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Trivnpk/Qwen-Compiler-Mind-4B:Q4_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 Trivnpk/Qwen-Compiler-Mind-4B:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use Trivnpk/Qwen-Compiler-Mind-4B with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Trivnpk/Qwen-Compiler-Mind-4B:Q4_K_M

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "Trivnpk/Qwen-Compiler-Mind-4B:Q4_K_M" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

Qwen-Compiler-Mind-4B v3.2.1 (PAL Arithmetic & Latent Static Analysis Edition)

Qwen-Compiler-Mind-4B v3.2.1 represents the definitive mathematical and algorithmic convergence of the Compiler Mind architecture. By embedding Program-Aided Language (PAL) Scratchpad Grounding and the Latent Static Analysis Invariant (LSAI), v3.2.1 achieves a massive leap on quantitative mathematical reasoning—catapulting GSM8K from 37.0% to an all-time record 78.0% (+41.0% absolute gain)—while solidly anchoring 80.0% Pass@1 on HumanEval and 55.5% on MMLU College CS & Formal Logic.

🏆 Multi-Generation Benchmark Scorecard (All 5 Generations)

Generation	HumanEval (Pass@1, 50s)	GSM8K (CoT, 100s)	MMLU (College CS & Formal Logic, 100s)	Cognitive Density $\rho$	Avg Thought Latency
v2.1	76.0%	52.0%	42.0%	0.90	1.8s
v3.0	78.0%	66.0%	50.0%	0.85	1.0s
v3.1.1 (Surgical Alignment)	82.0%	24.0% (Timeout)	25.0% (Random Guess Ceiling)	0.17 (Diluted)	14.2s
v3.2 (EPE Edition)	84.0%	37.0% (Mental Math Slip)	56.0% (Breakthrough)	0.62	5.7s
v3.2.1 (Current Release)	80.0% 🚀	78.0% 🏆 (All-Time High)	55.5% 🧠 (Logic Invariant)	0.65 (Optimal)	6.1s (Fluid)

Evaluated on NVIDIA L4 Tensor Core GPU using zero-dependency direct JSONL test runners.

🧠 Core Architectural Pillars in v3.2.1

1. Program-Aided Language (PAL) Scratchpad Grounding

The Problem Solved: Small 4B models frequently stumble when performing multi-digit mental arithmetic inside continuous token embeddings, leading to off-by-one errors on word problems ($14 \times 17 - 38$).
The PAL Invariant: Injects strict calculator scratchpad anchor tokens (<<expr = ans>>) directly into the reasoning flow:
```
Natalia sold 48/2 = <<48/2=24>>24 clips in May.
Natalia sold 48+24 = <<48+24=72>>72 clips altogether.
```
Result: GSM8K leaped from 37.0% straight to 78.0% (+41.0% absolute increase).

2. Latent Static Analysis Invariant (LSAI)

Before concluding thought and generating executable functions, the model verifies 3 non-negotiable compiler checks:

Type Safety Check: Guard against null pointers, empty collections, and signed integer overflow.
Off-by-One Guard: Strict bounds verification on loop and slice invariants (range(n) vs range(n + 1)).
Asymptotic Complexity Bound: Confirms optimal algorithmic time/space constraints.

3. Tri-Phase Elimination Protocol (TPEP)

Maintained the breakthrough on College Computer Science & Formal Logic (55.5%), completely preventing the 25% random-guess collapse of verbose CoT models.

4. Curriculum Structure (3,200 Pure Golden Samples)

1,100 Compiler & Systems Samples: C++20 SPSC queues, LLVM SSA dominance frontiers, AVX2 FMA kernels, Rust lock-free memory pools.
1,000 Formal Logic & CS Theory Samples: 949 real academic MMLU questions with counter-example falsification.
1,100 Real GSM8K Math Samples: Diverse real word problems with calculator scratchpads.
100% Zero Chinese tokens, 0% finance, 0% psychology chitchat.

📦 Quantized GGUF Triplet (Available Now)

Compiled natively with CMake AVX2/SIMD:

File Name	Precision	File Size	Recommended Target
`qwen-compiler-mind-v3.2.1-Q8_0.gguf`	8-bit	~4.48 GB	Reference accuracy, server inference, benchmarking
`qwen-compiler-mind-v3.2.1-Q6_K.gguf`	6-bit	~3.46 GB	Sweet spot: Laptop CPU / Intel Iris Xe (~22 tok/s)
`qwen-compiler-mind-v3.2.1-Q4_K_M.gguf`	4-bit	~2.59 GB	Low-memory edge devices, mobile RAM

🚀 Running with Ollama

# 1. Download Modelfile and Q6_K GGUF
ollama create qwen-compiler-mind-v3.2.1 -f Modelfile

# 2. Run interactively
ollama run qwen-compiler-mind-v3.2.1

📜 Citations & Architecture Credits

Architects: Ritmada (@RitmadaPrompt) & Tri (@Trivnpk)
Base Architecture: Qwen2.5-Coder-3B-Instruct / Continual Adaptation on Organic Reasoning Datasets
License: Apache 2.0

Downloads last month: 1,903

Safetensors

Model size

4B params

Tensor type

BF16

Model tree for Trivnpk/Qwen-Compiler-Mind-4B

Unable to build the model tree, the base model loops to the model itself. Learn more.