Skip to content

💻 CLI Commands Reference

Turing Engine provides a unified command-line tool turing for instant terminal chat, continuous batch serving, text generation, offline conversion, and accuracy evaluation.

Available Commands

1. turing chat

Start an interactive terminal chat with real pretrained weights: 

# Chat with SmolLM2 or DeepSeek-R1 in your terminal:
turing chat --model smollm2
turing chat --model deepseek-r1-1.5b
turing chat --model qwen3-coder-30b

2. turing serve

Start the continuous batching inference server with OpenAI and Anthropic-compatible endpoints: 

turing serve \
  --model deepseek-r1-7b \
  --port 8000 \
  --host 0.0.0.0 \
  --device auto \
  --sparsity 0.50 \
  --max-batch-size 32

3. turing generate

Generate text from a single prompt: 

turing generate --model gpt2 --prompt "Artificial intelligence is" --max-tokens 64

4. turing eval-accuracy

Run live mathematical reasoning (GSM8K) accuracy evaluation: 

turing eval-accuracy --model gpt2 --samples 10

5. turing bench

Run high-precision latency and throughput benchmarks: 

turing bench --model qwen3-coder-30b --batch-size 16 --seq-len 2048

6. turing convert

Convert HuggingFace safetensors checkpoints into Subspace-pruned format: 

turing convert --input ./llama-3.3-70b --output ./llama-3.3-70b-subspace --sparsity 0.57