qwen25-deposium-1024d / metadata.json
tss-deposium's picture
Upload metadata.json with huggingface_hub
c7f9792 verified
raw
history blame
976 Bytes
{
"base_model": "Qwen/Qwen2.5-1.5B-Instruct",
"model_size": "1.54B parameters",
"model_type": "instruction-tuned LLM",
"dimensions": 1024,
"model2vec_version": "0.7.0",
"custom_corpus": false,
"note": "First-ever Model2Vec distillation of instruction-tuned LLM",
"output_dir": "models/qwen25-deposium-1024d",
"expected_quality": "0.75-0.85+",
"expected_speedup": "500-1000x vs full LLM",
"final_size_mb": 65,
"multilingual": true,
"instruction_aware": true,
"context_length": 32768,
"unique_capabilities": [
"Instruction-aware embeddings",
"Superior semantic understanding",
"Conversational context",
"Code understanding",
"Idiom recognition"
],
"advantages_vs_competitors": {
"vs_qwen3_embedding": "10x smaller, instruction-aware",
"vs_gemma_768d": "larger base model, instruction-tuned",
"vs_qwen3_256d": "higher quality expected, instruction-aware"
},
"instruction_awareness_score": 0.9198921983524119
}