Upload encoding/README.md with huggingface_hub
Browse files- encoding/README.md +22 -0
encoding/README.md
ADDED
|
@@ -0,0 +1,22 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# DeepSeek V3.2 Message Encoding
|
| 2 |
+
|
| 3 |
+
This directory contains the message encoding implementation and tests for DeepSeek V3.2.
|
| 4 |
+
|
| 5 |
+
## Files
|
| 6 |
+
|
| 7 |
+
- `encoding_dsv32.py` - DeepSeek V3.2 message encoding implementation
|
| 8 |
+
- `test_encoding_dsv32.py` - Unit tests for encoding functionality
|
| 9 |
+
- `test_input*.json` - Test cases for encoding validation
|
| 10 |
+
|
| 11 |
+
## Usage
|
| 12 |
+
|
| 13 |
+
The encoding module is used by the inference implementation to format messages with proper special tokens:
|
| 14 |
+
|
| 15 |
+
```python
|
| 16 |
+
from encoding_dsv32 import encode_messages
|
| 17 |
+
|
| 18 |
+
messages = [{"role": "user", "content": "Hello"}]
|
| 19 |
+
prompt = encode_messages(messages, thinking_mode="chat")
|
| 20 |
+
```
|
| 21 |
+
|
| 22 |
+
See `../inference/README.md` for usage in the full inference pipeline.
|