Rename BankAI training configuration
Browse files- training_config.yaml +2 -2
training_config.yaml
CHANGED
|
@@ -1,7 +1,7 @@
|
|
| 1 |
# QLoRA for Qwen3-Coder-Next on Apple Silicon with 128 GB unified memory.
|
| 2 |
model: mlx-community/Qwen3-Coder-Next-4bit
|
| 3 |
train: true
|
| 4 |
-
data: artifacts/
|
| 5 |
fine_tune_type: lora
|
| 6 |
optimizer: adamw
|
| 7 |
mask_prompt: true
|
|
@@ -14,7 +14,7 @@ learning_rate: 1.0e-5
|
|
| 14 |
steps_per_report: 5
|
| 15 |
steps_per_eval: 100
|
| 16 |
save_every: 100
|
| 17 |
-
adapter_path: artifacts/
|
| 18 |
max_seq_length: 2048
|
| 19 |
grad_checkpoint: true
|
| 20 |
clear_cache_threshold: 68719476736
|
|
|
|
| 1 |
# QLoRA for Qwen3-Coder-Next on Apple Silicon with 128 GB unified memory.
|
| 2 |
model: mlx-community/Qwen3-Coder-Next-4bit
|
| 3 |
train: true
|
| 4 |
+
data: artifacts/bankai-v1/data
|
| 5 |
fine_tune_type: lora
|
| 6 |
optimizer: adamw
|
| 7 |
mask_prompt: true
|
|
|
|
| 14 |
steps_per_report: 5
|
| 15 |
steps_per_eval: 100
|
| 16 |
save_every: 100
|
| 17 |
+
adapter_path: artifacts/bankai-v1/adapter
|
| 18 |
max_seq_length: 2048
|
| 19 |
grad_checkpoint: true
|
| 20 |
clear_cache_threshold: 68719476736
|