[Bug] Not gettting of output_dir from train_kwargs in dsp/dspy/clients/lm_local.py

[rubensmau]

What happened?

When running https://dspy.ai/tutorials/classification_finetuning/ inside Google Colab, Line:
classify_ft = optimizer.compile(student_classify, teacher=teacher_classify, trainset=unlabeled_trainset)

got error:

[BootstrapFinetune] Preparing the student and teacher programs...
[BootstrapFinetune] Bootstrapping data...
Average Metric: 500.00 / 500 (100.0%): 100%|██████████| 500/500 [00:04<00:00, 115.30it/s]2025/02/05 21:40:32 INFO dspy.evaluate.evaluate: Average Metric: 500 / 500 (100.0%)

[BootstrapFinetune] Preparing the train data...
[BootstrapFinetune] Using 500 data points for fine-tuning the model: openai/local:meta-llama/Llama-3.2-1B-Instruct
[BootstrapFinetune] Starting LM fine-tuning...
[BootstrapFinetune] 1 fine-tuning job(s) to start
[BootstrapFinetune] Starting 1 fine-tuning job(s)...
[Local Provider] Data saved to /root/.dspy_cache/finetune/4bec8714ae43838b.jsonl
[Local Provider] Starting local training, will save to /root/.dspy_cache/finetune/4bec8714ae43838b__meta-llama__Llama-3.2-1B-Instruct
Using device: cuda
Adding pad token to tokenizer
Creating dataset
Map: 100%
 500/500 [00:00<00:00, 2403.78 examples/s]
2025/02/05 21:40:57 ERROR dspy.clients.lm: name 'output_dir' is not defined

It seems to me that the problem is in dsp/dspy/clients/lm_local.py
it is missing in the code, in line 180:
output_dir = train_kwargs.get("output_dir", None)

Steps to reproduce

run notebook

DSPy version

2.6.2

[okhat]

Hey yes we have a PR #7629 !