docs: add portable usage and training config link
Browse files
README.md
CHANGED
|
@@ -47,7 +47,9 @@ This repository contains an adapter only. It requires the base model above and a
|
|
| 47 |
from mlx_lm import generate, load
|
| 48 |
|
| 49 |
base_model = "mlx-community/Qwen2.5-Coder-3B-Instruct-4bit"
|
| 50 |
-
|
|
|
|
|
|
|
| 51 |
|
| 52 |
model, tokenizer = load(base_model, adapter_path=adapter_path)
|
| 53 |
messages = [
|
|
@@ -72,6 +74,9 @@ print(generate(model, tokenizer, prompt=prompt, max_tokens=256, verbose=False))
|
|
| 72 |
- Split: 5,000 training and 1,000 held-out rows grouped by identical full schema context; zero shared context groups
|
| 73 |
- Training: 2,500 micro-batches, batch size 2, gradient accumulation 4, learning rate `1e-5`, prompt loss masking
|
| 74 |
|
|
|
|
|
|
|
|
|
|
| 75 |
## Limitations
|
| 76 |
|
| 77 |
- This is a local, schema-conditioned subset experiment, not a production SQL agent.
|
|
|
|
| 47 |
from mlx_lm import generate, load
|
| 48 |
|
| 49 |
base_model = "mlx-community/Qwen2.5-Coder-3B-Instruct-4bit"
|
| 50 |
+
# Replace this placeholder with the local directory containing the downloaded
|
| 51 |
+
# files from this Hugging Face repository.
|
| 52 |
+
adapter_path = "/path/to/qwen25-coder-3b-sql-create-context-lora-mlx"
|
| 53 |
|
| 54 |
model, tokenizer = load(base_model, adapter_path=adapter_path)
|
| 55 |
messages = [
|
|
|
|
| 74 |
- Split: 5,000 training and 1,000 held-out rows grouped by identical full schema context; zero shared context groups
|
| 75 |
- Training: 2,500 micro-batches, batch size 2, gradient accumulation 4, learning rate `1e-5`, prompt loss masking
|
| 76 |
|
| 77 |
+
The complete, runnable training recipe—including project-relative data and
|
| 78 |
+
adapter paths—is [`configs/lora.yaml`](https://github.com/karyboy/mlx-text-to-sql-lora/blob/main/configs/lora.yaml).
|
| 79 |
+
|
| 80 |
## Limitations
|
| 81 |
|
| 82 |
- This is a local, schema-conditioned subset experiment, not a production SQL agent.
|