Upload folder using huggingface_hub

HybriKo_tok.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8a9651005063f8bf9efc66d7333da8e99f72dba48791e35d57429159c2f891bb
+size 805880

README.md

@@ -0,0 +1,88 @@
+# HybriKo-117M-LinuxFC-SFT-v2
+
+Korean Hybrid LLM fine-tuned for Linux Command Function Calling.
+
+## Model Description
+
+- **Architecture**: Griffin-style Hybrid (RNN + Attention, 2:1 ratio)
+- **Parameters**: 117.8M
+- **Base Model**: HybriKo-117M (exp7_phase1)
+- **Fine-tuning**: Linux Function Calling SFT
+- **Training Data**: 4,725 samples (21 Linux commands)
+
+## Performance
+
+| Metric | Value |
+|--------|-------|
+| **Action Name Accuracy** | **100%** (100/100) |
+| Eval Loss | 0.0039 |
+| Training Epochs | 15 |
+
+## Supported Commands (21)
+
+`ls`, `cd`, `mkdir`, `rm`, `cp`, `mv`, `find`, `cat`, `grep`, `head`, `tail`, `wc`, `ps`, `df`, `du`, `top`, `ping`, `curl`, `chmod`, `tar`, `Finish`
+
+## Usage
+
+```python
+import torch
+import sentencepiece as spm
+from hybridko.model import HybriKoModel, HybriKoConfig
+
+# Load tokenizer
+sp = spm.SentencePieceProcessor()
+sp.Load("HybriKo_tok.model")
+
+# Load model
+config = HybriKoConfig(
+    d_model=768, n_layers=12, vocab_size=32000,
+    n_heads=12, n_kv_heads=3, ff_mult=3,
+    max_seq_len=6144, dropout=0.0
+)
+model = HybriKoModel(config)
+checkpoint = torch.load("pytorch_model.pt", map_location="cpu")
+model.load_state_dict(checkpoint["model_state_dict"])
+model.eval()
+
+# Inference
+prompt = """<|im_start|>system
+You are a Linux command assistant.
+<|im_end|>
+<|im_start|>user
+현재 폴더의 파일 목록을 보여줘
+<|im_end|>
+<|im_start|>assistant
+"""
+# Generate response...
+```
+
+## Output Format
+
+```
+Thought: 디렉토리 내용을 확인합니다.
+Action: ls_command
+Action Input: {"path": ".", "options": "-l"}
+```
+
+## Training Details
+
+- **Hardware**: A100 x 8 (DDP)
+- **Batch Size**: 32 (1 per GPU x 8 GPUs x 4 grad accum)
+- **Learning Rate**: 5e-5
+- **Warmup Steps**: 100
+- **Epochs**: 15 (converged)
+
+## License
+
+Apache 2.0
+
+## Citation
+
+```bibtex
+@misc{hybridko-linuxfc-2026,
+  title={HybriKo-117M-LinuxFC-SFT-v2},
+  author={Yaongi},
+  year={2026},
+  publisher={HuggingFace}
+}
+```

config.yaml

@@ -0,0 +1,31 @@
+# HybriKo Default Configuration
+# ~117.8M parameters, optimized for Colab T4 (16GB VRAM)
+
+model:
+  d_model: 768        # Hidden dimension
+  n_layers: 12        # Number of hybrid layers
+  vocab_size: 32000   # Vocabulary size
+  n_heads: 12         # Attention heads
+  n_kv_heads: 3       # KV heads for GQA (1:4 ratio)
+  ff_mult: 3          # Feed-forward multiplier
+  max_seq_len: 1024   # Maximum sequence length
+
+training:
+  learning_rate: 3.0e-4
+  weight_decay: 0.1
+  warmup_steps: 20
+  max_steps: 1000
+  grad_accum_steps: 1
+  save_steps: 500
+  batch_size: 16
+  max_length: 512     # Training sequence length
+
+data:
+  num_samples: 30000
+  min_length: 50
+  tokenizer_samples: 100000
+
+tokenizer:
+  vocab_size: 32000
+  model_type: unigram
+  character_coverage: 0.9995

pytorch_model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:222a76320a48bfdb5b9673ba5926cb6be7274b3a294db50bc52577caa5cda6f7
+size 1414061818