Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,84 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: cc-by-nc-4.0
|
| 3 |
+
datasets:
|
| 4 |
+
- iamtarun/python_code_instructions_18k_alpaca
|
| 5 |
+
language:
|
| 6 |
+
- en
|
| 7 |
+
base_model: Pacific-Prime/pacific-prime-code
|
| 8 |
+
tags:
|
| 9 |
+
- code
|
| 10 |
+
- python
|
| 11 |
+
- i64
|
| 12 |
+
- complexity-deep
|
| 13 |
+
- sft
|
| 14 |
+
pipeline_tag: text-generation
|
| 15 |
+
library_name: transformers
|
| 16 |
+
---
|
| 17 |
+
|
| 18 |
+
# Pacific-Prime: Python Node
|
| 19 |
+
|
| 20 |
+
**Pure Python specialist** fine-tuned from Pacific-Prime Code (I64 architecture, 1.5B parameters).
|
| 21 |
+
|
| 22 |
+
## Skills
|
| 23 |
+
|
| 24 |
+
- Python basics & standard library
|
| 25 |
+
- Algorithms & data structures
|
| 26 |
+
- Object-oriented programming
|
| 27 |
+
- Decorators & generators
|
| 28 |
+
- List comprehensions
|
| 29 |
+
- File I/O & error handling
|
| 30 |
+
|
| 31 |
+
## Training
|
| 32 |
+
|
| 33 |
+
- **Architecture**: I64 (Complexity-Deep)
|
| 34 |
+
- **Parameters**: 1.5B
|
| 35 |
+
- **Base model**: pacific-prime-code (checkpoint epoch 70)
|
| 36 |
+
- **Method**: Full SFT (no LoRA)
|
| 37 |
+
- **Dataset**: python_code_instructions_18k_alpaca (18K samples)
|
| 38 |
+
- **Epochs**: 1000
|
| 39 |
+
- **Max context**: 4096 tokens
|
| 40 |
+
|
| 41 |
+
## Inference with vLLM-I64
|
| 42 |
+
|
| 43 |
+
Use our custom vLLM engine with native I64 support:
|
| 44 |
+
|
| 45 |
+
**[vllm-i64](https://github.com/Complexity-ML/vllm-i64)**
|
| 46 |
+
|
| 47 |
+
```bash
|
| 48 |
+
git clone https://github.com/Complexity-ML/vllm-i64.git
|
| 49 |
+
cd vllm-i64
|
| 50 |
+
pip install -e .
|
| 51 |
+
```
|
| 52 |
+
|
| 53 |
+
```python
|
| 54 |
+
from vllm import LLM, SamplingParams
|
| 55 |
+
|
| 56 |
+
model = LLM(model="Pacific-Prime/python-node")
|
| 57 |
+
params = SamplingParams(temperature=0.7, max_tokens=4096)
|
| 58 |
+
|
| 59 |
+
prompt = "User: Write a Python function to find the longest common subsequence of two strings.\nAssistant:"
|
| 60 |
+
output = model.generate([prompt], params)
|
| 61 |
+
print(output[0].outputs[0].text)
|
| 62 |
+
```
|
| 63 |
+
|
| 64 |
+
## Serve Your Own I64 Model
|
| 65 |
+
|
| 66 |
+
Trained your own I64 model with [complexity-deep](https://github.com/Complexity-ML/complexity-deep)? Serve it with vllm-i64:
|
| 67 |
+
|
| 68 |
+
```python
|
| 69 |
+
from vllm import LLM, SamplingParams
|
| 70 |
+
|
| 71 |
+
model = LLM(model="/path/to/your/i64-model")
|
| 72 |
+
params = SamplingParams(temperature=0.7, max_tokens=4096)
|
| 73 |
+
output = model.generate(["User: Hello!\nAssistant:"], params)
|
| 74 |
+
print(output[0].outputs[0].text)
|
| 75 |
+
```
|
| 76 |
+
|
| 77 |
+
## Links
|
| 78 |
+
|
| 79 |
+
- [Complexity-Deep](https://github.com/Complexity-ML/complexity-deep) — Training framework & architecture
|
| 80 |
+
- [vllm-i64](https://github.com/Complexity-ML/vllm-i64) — Inference engine for I64 models
|
| 81 |
+
|
| 82 |
+
## License
|
| 83 |
+
|
| 84 |
+
CC BY-NC 4.0
|