Skip to content

Fix LoRA reload path in Gemma4 and Qwen3.5 GRPO notebooks #291

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
danielhanchen wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Fix LoRA reload path in Gemma4 and Qwen3.5 GRPO notebooks #291

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
4b35104
Whisper: pass processing_class instead of removed tokenizer kwarg
danielhanchen Jun 12, 2026
b2ad979
Fix LoRA reload path in Gemma4 and Qwen3.5 GRPO notebooks
danielhanchen Jun 13, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion nb/Gemma4_(E2B)_GRPO.ipynb
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5364,7 +5364,7 @@
"from safetensors import safe_open\n",
"\n",
"tensors = {}\n",
"with safe_open(\"grpo_saved_lora/adapter_model.safetensors\", framework = \"pt\") as f:\n",
"with safe_open(\"gemma_4_lora/adapter_model.safetensors\", framework = \"pt\") as f:\n",
" # Verify both A and B are non zero\n",
" for key in f.keys():\n",
" tensor = f.get_tensor(key)\n",
Expand Down
2 changes: 1 addition & 1 deletion nb/Gemma4_(E2B)_Reinforcement_Learning_2048_Game.ipynb
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11731,7 +11731,7 @@
"from safetensors import safe_open\n",
"\n",
"tensors = {}\n",
"with safe_open(\"grpo_saved_lora/adapter_model.safetensors\", framework = \"pt\") as f:\n",
"with safe_open(\"gemma_4_lora/adapter_model.safetensors\", framework = \"pt\") as f:\n",
" # Verify both A and B are non zero\n",
" for key in f.keys():\n",
" tensor = f.get_tensor(key)\n",
Expand Down
2 changes: 1 addition & 1 deletion nb/Gemma4_(E2B)_Reinforcement_Learning_Sudoku_Game.ipynb
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8163,7 +8163,7 @@
"from safetensors import safe_open\n",
"\n",
"tensors = {}\n",
"with safe_open(\"grpo_saved_lora/adapter_model.safetensors\", framework = \"pt\") as f:\n",
"with safe_open(\"gemma_4_lora/adapter_model.safetensors\", framework = \"pt\") as f:\n",
" # Verify both A and B are non zero\n",
" for key in f.keys():\n",
" tensor = f.get_tensor(key)\n",
Expand Down
2 changes: 1 addition & 1 deletion nb/Kaggle-Whisper.ipynb
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -675,7 +675,7 @@
" train_dataset = train_dataset,\n",
" data_collator = DataCollatorSpeechSeq2SeqWithPadding(processor = tokenizer),\n",
" eval_dataset = test_dataset,\n",
" tokenizer = tokenizer.feature_extractor,\n",
" processing_class = tokenizer.feature_extractor,\n",

@gemini-code-assist gemini-code-assist Bot Jun 13, 2026

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

medium

Since tokenizer is actually a WhisperProcessor (as indicated by its usage in DataCollatorSpeechSeq2SeqWithPadding(processor = tokenizer)), you can pass the processor itself directly to processing_class instead of just its feature extractor. This is the recommended approach in modern Transformers, and it ensures that both the feature extractor and the tokenizer are available to the trainer (e.g., for saving the processor or if predict_with_generate is enabled).

Suggested change
" processing_class = tokenizer.feature_extractor,\n",
" processing_class = tokenizer,\n",

" compute_metrics = compute_metrics,\n",
" args = Seq2SeqTrainingArguments(\n",
" # predict_with_generate = True,\n",
Expand Down
4 changes: 2 additions & 2 deletions nb/Qwen3_5_(4B)_Vision_GRPO.ipynb
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8435,7 +8435,7 @@
"output_type": "execute_result",
"data": {
"text/plain": [
"['grpo_lora/processor_config.json']"
"['qwen_lora/processor_config.json']"
]
},
"metadata": {},
Expand Down Expand Up @@ -8469,7 +8469,7 @@
"from safetensors import safe_open\n",
"\n",
"tensors = {}\n",
"with safe_open(\"grpo_lora/adapter_model.safetensors\", framework = \"pt\") as f:\n",
"with safe_open(\"qwen_lora/adapter_model.safetensors\", framework = \"pt\") as f:\n",
" # Verify both A and B are non zero\n",
" for key in f.keys():\n",
" tensor = f.get_tensor(key)\n",
Expand Down
2 changes: 1 addition & 1 deletion nb/Whisper.ipynb
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -675,7 +675,7 @@
" train_dataset = train_dataset,\n",
" data_collator = DataCollatorSpeechSeq2SeqWithPadding(processor = tokenizer),\n",
" eval_dataset = test_dataset,\n",
" tokenizer = tokenizer.feature_extractor,\n",
" processing_class = tokenizer.feature_extractor,\n",

@gemini-code-assist gemini-code-assist Bot Jun 13, 2026

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

medium

Since tokenizer is actually a WhisperProcessor (as indicated by its usage in DataCollatorSpeechSeq2SeqWithPadding(processor = tokenizer)), you can pass the processor itself directly to processing_class instead of just its feature extractor. This is the recommended approach in modern Transformers, and it ensures that both the feature extractor and the tokenizer are available to the trainer (e.g., for saving the processor or if predict_with_generate is enabled).

Suggested change
" processing_class = tokenizer.feature_extractor,\n",
" processing_class = tokenizer,\n",

" compute_metrics = compute_metrics,\n",
" args = Seq2SeqTrainingArguments(\n",
" # predict_with_generate = True,\n",
Expand Down
2 changes: 1 addition & 1 deletion original_template/Whisper.ipynb
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -682,7 +682,7 @@
" train_dataset = train_dataset,\n",
" data_collator = DataCollatorSpeechSeq2SeqWithPadding(processor=tokenizer),\n",
" eval_dataset = test_dataset,\n",
" tokenizer = tokenizer.feature_extractor,\n",
" processing_class = tokenizer.feature_extractor,\n",

@gemini-code-assist gemini-code-assist Bot Jun 13, 2026

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

medium

Since tokenizer is actually a WhisperProcessor (as indicated by its usage in DataCollatorSpeechSeq2SeqWithPadding(processor=tokenizer)), you can pass the processor itself directly to processing_class instead of just its feature extractor. This is the recommended approach in modern Transformers, and it ensures that both the feature extractor and the tokenizer are available to the trainer (e.g., for saving the processor or if predict_with_generate is enabled).

Suggested change
" processing_class = tokenizer.feature_extractor,\n",
" processing_class = tokenizer,\n",

" compute_metrics=compute_metrics,\n",
" args = Seq2SeqTrainingArguments(\n",
" # predict_with_generate=True,\n",
Expand Down
2 changes: 1 addition & 1 deletion python_scripts/Kaggle-Whisper.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -195,7 +195,7 @@ def __call__(self, features: List[Dict[str, Union[List[int], torch.Tensor]]]) ->
train_dataset = train_dataset,
data_collator = DataCollatorSpeechSeq2SeqWithPadding(processor = tokenizer),
eval_dataset = test_dataset,
tokenizer = tokenizer.feature_extractor,
processing_class = tokenizer.feature_extractor,

@gemini-code-assist gemini-code-assist Bot Jun 13, 2026

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

medium

Since tokenizer is actually a WhisperProcessor (as indicated by its usage in DataCollatorSpeechSeq2SeqWithPadding(processor = tokenizer)), you can pass the processor itself directly to processing_class instead of just its feature extractor. This is the recommended approach in modern Transformers, and it ensures that both the feature extractor and the tokenizer are available to the trainer (e.g., for saving the processor or if predict_with_generate is enabled).

Suggested change
processing_class = tokenizer.feature_extractor,
processing_class = tokenizer,

compute_metrics = compute_metrics,
args = Seq2SeqTrainingArguments(
# predict_with_generate = True,
Expand Down
2 changes: 1 addition & 1 deletion python_scripts/Whisper.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -195,7 +195,7 @@ def __call__(self, features: List[Dict[str, Union[List[int], torch.Tensor]]]) ->
train_dataset = train_dataset,
data_collator = DataCollatorSpeechSeq2SeqWithPadding(processor = tokenizer),
eval_dataset = test_dataset,
tokenizer = tokenizer.feature_extractor,
processing_class = tokenizer.feature_extractor,

@gemini-code-assist gemini-code-assist Bot Jun 13, 2026

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

medium

Since tokenizer is actually a WhisperProcessor (as indicated by its usage in DataCollatorSpeechSeq2SeqWithPadding(processor = tokenizer)), you can pass the processor itself directly to processing_class instead of just its feature extractor. This is the recommended approach in modern Transformers, and it ensures that both the feature extractor and the tokenizer are available to the trainer (e.g., for saving the processor or if predict_with_generate is enabled).

Suggested change
processing_class = tokenizer.feature_extractor,
processing_class = tokenizer,

compute_metrics = compute_metrics,
args = Seq2SeqTrainingArguments(
# predict_with_generate = True,
Expand Down
Loading