GilesL/anki-card-generator-v3-5-gguf

Anki Card Generator V3.5 (GGUF)

This is a GGUF conversion of GilesL/anki-card-generator-v3-5, a fine-tuned model for generating Anki flashcards from text.

Model Details

• Base Model: Qwen/Qwen2.5-3B
• Fine-tuning: LoRA (r=32, alpha=64)
• Training Data: 90 premium examples (50 Claude Sonnet 4.5 + 40 Claude Opus 4.5)
• Training: 15 epochs
• Format: GGUF (for llama.cpp, Ollama, LM Studio, etc.)

What It Does

Automatically generates atomic Anki flashcards from raw text input. No prompt engineering needed - just provide the text and it generates structured JSON flashcards with front, back, and tags.

Available Quantizations

File	Quant	Size	Description	Use Case
anki-card-generator-v3-5-f16.gguf	F16	~6GB	Full precision	Best quality
anki-card-generator-v3-5-q8_0.gguf	Q8_0	~3GB	8-bit	High quality
anki-card-generator-v3-5-q5_k_m.gguf	Q5_K_M	~2GB	5-bit medium	Good quality
anki-card-generator-v3-5-q4_k_m.gguf	Q4_K_M	~1.7GB	4-bit medium	Recommended

Usage with Ollama

1. Download the model:

huggingface-cli download GilesL/anki-card-generator-v3-5-gguf anki-card-generator-v3-5-q4_k_m.gguf --local-dir .

2. Create a Modelfile:

FROM ./anki-card-generator-v3-5-q4_k_m.gguf
TEMPLATE """{{ .Prompt }}"""
PARAMETER temperature 0.7
PARAMETER top_p 0.9

3. Create and run:

ollama create anki-generator -f Modelfile
echo "Your text here" | ollama run anki-generator

Example Output

Input text about transformers generates:

[
  {
    "front": "What year was the transformer architecture introduced?",
    "back": "2017.",
    "tags": ["transformer", "history", "neural-networks"]
  },
  // ... more cards
]

License

Inherits license from Qwen/Qwen2.5-3B

---

Converted to GGUF using llama.cpp