Skip to content

TESTING.md

Latest commit

 

History

History
112 lines (79 loc) · 2.63 KB

TESTING.md

File metadata and controls

112 lines (79 loc) · 2.63 KB
Raw

Testing Guide

Quick Test

Verify the MinT SDK is working correctly:

# Set credentials
export MINT_BASE_URL=https://mint-training-api.macaron.xin
export MINT_API_KEY=your-api-key

# Run example
python examples/basic_training_setup.py

Expected output:

MinT SDK Example - v0.2.0
============================================================
Endpoint: https://mint-training-api.macaron.xin
API Key: your-api-key...

1. Creating ServiceClient...
   ✓ ServiceClient created

2. Getting available models...
   ✓ Found 1 model(s):
     - Qwen/Qwen3.6-35B-A3B (context: 65536)

3. Creating LoRA TrainingClient...
   Model: Qwen/Qwen3.6-35B-A3B
   LoRA rank: 16
   ✓ TrainingClient created

4. Loading tokenizer...
   ✓ Tokenizer ready: Qwen2Tokenizer
   Example: 'Hello, world!' → 4 tokens

============================================================
✓ Setup complete!

Test Results (2026-08-13)

Tested Endpoints

Endpoint Status Notes
https://mint-training-api.macaron.xin ✓ Working Recommended endpoint
https://mintcn.macaron.xin/train ✓ Working Alternative endpoint
https://mint-cn.macaron.xin/ ⚠️ Auth issues Key registration required

Tested Functionality

  • ✓ ServiceClient creation
  • ✓ API authentication (both sk- and tml- prefixes)
  • ✓ Server capabilities query
  • ✓ Model list retrieval
  • ✓ TrainingClient creation
  • ✓ Tokenizer loading

Known Working Models

  • Qwen/Qwen3.6-35B-A3B (context: 65536)

Use get_server_capabilities() to check currently available models on your endpoint.

Installation

# Create virtual environment
python3.12 -m venv .venv
source .venv/bin/activate

# Install dependencies
pip install -e .

# Verify installation
python -c "import mint; print(f'mint {mint.__version__}')"

API Key Formats

The SDK supports both key formats:

  • sk-... (MinT standard keys)
  • tml-... (Tinker/legacy keys)

Troubleshooting

AuthenticationError: "must start with the 'tml-' prefix"

This was a bug in v0.2.0 that has been fixed. The auth patch now handles both code paths:

  1. resolve_auth_provider (JWT-enabled flow)
  2. Direct ApiKeyAuthProvider instantiation (non-JWT flow)

Update to the latest version to resolve this issue.

TrainingClient creation timeout

Check:

  1. Your API key has training permissions
  2. The model name matches an available model from get_server_capabilities()
  3. Your account has sufficient quota/balance

Empty supported_models list

The endpoint may require:

  • Account activation
  • Billing setup
  • Different base URL

Contact MinT support for assistance.