Tambah section Identity reinforcement (v2: 30 contoh identity, 2 epoch, lr 1e-5)
Browse files
README.md
CHANGED
|
@@ -114,6 +114,12 @@ Kamu adalah GarudaCoder, asisten coding berbahasa Indonesia yang teliti. Utamaka
|
|
| 114 |
- Holdout: 80 examples kept out of training entirely
|
| 115 |
- Anti-leakage: holdout/benchmark files gated (FILE_DILARANG_UNTUK_TRAINING)
|
| 116 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 117 |
## Lineage
|
| 118 |
- Predecessor: [GarudaCoder-7B-Coder-ID-lora](https://huggingface.co/TheCoderScientist/GarudaCoder-7B-Coder-ID-lora) (Qwen2.5-Coder-7B)
|
| 119 |
- This qwen3_5 model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth)
|
|
|
|
| 114 |
- Holdout: 80 examples kept out of training entirely
|
| 115 |
- Anti-leakage: holdout/benchmark files gated (FILE_DILARANG_UNTUK_TRAINING)
|
| 116 |
|
| 117 |
+
|
| 118 |
+
## Identity reinforcement
|
| 119 |
+
- v2 menambahkan 30 contoh identity bilingual (perkenalan, provokasi "kamu Qwen", capabilities, out-of-scope refusal, coding+self-reference, "cuma program") di atas data training. Tujuannya: identity GarudaCoder tetap claim ketika user provokasi ("kamu kan Qwen?"), bukan fallback ke base.
|
| 120 |
+
- Anti-bocor: tidak ada penyebutan sumber training eksternal di data identity.
|
| 121 |
+
- v2 dilatih 2 epoch, lr 1e-5 (v1: 1 epoch, lr 2e-5). Adapter v2 replace v1 di repo ini.
|
| 122 |
+
|
| 123 |
## Lineage
|
| 124 |
- Predecessor: [GarudaCoder-7B-Coder-ID-lora](https://huggingface.co/TheCoderScientist/GarudaCoder-7B-Coder-ID-lora) (Qwen2.5-Coder-7B)
|
| 125 |
- This qwen3_5 model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth)
|