Upload metadata.json with huggingface_hub
Browse files- metadata.json +29 -0
metadata.json
ADDED
|
@@ -0,0 +1,29 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"base_model": "Qwen/Qwen2.5-1.5B-Instruct",
|
| 3 |
+
"model_size": "1.54B parameters",
|
| 4 |
+
"model_type": "instruction-tuned LLM",
|
| 5 |
+
"dimensions": 1024,
|
| 6 |
+
"model2vec_version": "0.7.0",
|
| 7 |
+
"custom_corpus": false,
|
| 8 |
+
"note": "First-ever Model2Vec distillation of instruction-tuned LLM",
|
| 9 |
+
"output_dir": "models/qwen25-deposium-1024d",
|
| 10 |
+
"expected_quality": "0.75-0.85+",
|
| 11 |
+
"expected_speedup": "500-1000x vs full LLM",
|
| 12 |
+
"final_size_mb": 65,
|
| 13 |
+
"multilingual": true,
|
| 14 |
+
"instruction_aware": true,
|
| 15 |
+
"context_length": 32768,
|
| 16 |
+
"unique_capabilities": [
|
| 17 |
+
"Instruction-aware embeddings",
|
| 18 |
+
"Superior semantic understanding",
|
| 19 |
+
"Conversational context",
|
| 20 |
+
"Code understanding",
|
| 21 |
+
"Idiom recognition"
|
| 22 |
+
],
|
| 23 |
+
"advantages_vs_competitors": {
|
| 24 |
+
"vs_qwen3_embedding": "10x smaller, instruction-aware",
|
| 25 |
+
"vs_gemma_768d": "larger base model, instruction-tuned",
|
| 26 |
+
"vs_qwen3_256d": "higher quality expected, instruction-aware"
|
| 27 |
+
},
|
| 28 |
+
"instruction_awareness_score": 0.9198921983524119
|
| 29 |
+
}
|