UnboundLocalError on Arm64 M3 Mac with blablador backend

[thawn]

Whenever I submit a prompt to blablador, I get the following error message:

UnboundLocalError: cannot access local variable 'instructions' where it is not associated with a value

---------------------------------------------------------------------------
UnboundLocalError                         Traceback (most recent call last)
Cell In[2], line 1
----> 1 get_ipython().run_cell_magic('bob', '', 'load blobs.tif and show it\n')
File /opt/homebrew/Caskroom/mambaforge/base/envs/bob-env/lib/python3.11/site-packages/IPython/core/interactiveshell.py:2541, in InteractiveShell.run_cell_magic(self, magic_name, line, cell)

2539 with self.builtin_trap:

2540     args = (magic_arg_s, cell)

-> 2541     result = fn(*args, **kwargs)

2543 # The code below prevents the output from being displayed

2544 # when using magics with decorator @output_can_be_silenced

2545 # when the last Python token in the expression is a ';'.

2546 if getattr(fn, magic.MAGIC_OUTPUT_CAN_BE_SILENCED, False):
File /opt/homebrew/Caskroom/mambaforge/base/envs/bob-env/lib/python3.11/site-packages/bia_bob/_machinery.py:140, in bob(line, cell)

138         text = f"The file has been saved as {filename}. You can open it using:\n\n    jupyter lab {filename}\n\n"

139 else:

--> 140     code, text = generate_response_to_user(Context.model, user_input, image)

142     if image is not None:

143         # we need to add this information to the history.

144         Context.chat.append({

145             "role": "user",

146             "content": f"Assume there is an image stored in variable {image_name}. The image can be described like this: {response}. Just confirm this with 'ok'."

147         })
File /opt/homebrew/Caskroom/mambaforge/base/envs/bob-env/lib/python3.11/site-packages/bia_bob/_utilities.py:27, in generate_response_to_user(model, user_prompt, image, additional_system_prompt, max_number_attempts, system_prompt)

25 for attempt in range(1, max_number_attempts + 1):

26     if system_prompt is None:

---> 27         system_prompt = create_system_prompt()

28     if additional_system_prompt is not None:

29         system_prompt += "\n" + additional_system_prompt
File /opt/homebrew/Caskroom/mambaforge/base/envs/bob-env/lib/python3.11/site-packages/bia_bob/_utilities.py:231, in create_system_prompt(reusable_variables_block)

228 else:

229     warnings.warn("Deprecated use of create_system_prompt with reusable_variables_block. Do not pass this parameter to make your code work mid/long-term.")

--> 231 snippets, additional_snippets = generate_code_samples()

232 libraries = Context.libraries

234 system_prompt = f"""

235 You are a extremely talented bioimage analyst and you use Python to solve your tasks unless stated otherwise.

236 If there are several ways to solve the task, chose the option with the least amount of code.

(...)

278 Make sure to keep your answer concise and to the point. Make sure the code you write is correct and can be executed.

279 """
File /opt/homebrew/Caskroom/mambaforge/base/envs/bob-env/lib/python3.11/site-packages/bia_bob/_utilities.py:211, in generate_code_samples()

209         pass

210     # special treatment for code snippets from stackview, as it won't work with the custom kernel

--> 211     if "stackview" not in instructions or "stackview" in Context.libraries:

212         additional_instructions.append(instructions)

214 additional_snippets = "\n".join(additional_instructions)
UnboundLocalError: cannot access local variable 'instructions' where it is not associated with a value

minimal notebook:

biabob.ipynb.zip

steps to reproduce:

mamba env create -f https://github.com/haesleinhuepf/bia-bob/raw/main/environment.yml
mamba activate bob-env
mamba env config vars set BLABLADOR_API_KEY=<my API key>
mamba activate bob-env #to enable the BLABLADOR_API_KEY env variable
jupyter lab

[haesleinhuepf]

Hi @thawn ,

thanks for the report. Unfortunately, I cannot reproduce this on my PC:

As a hardware-related issue is unlikely, I presume it has something to do with your conda environment. Could you share the output of conda list?

Thanks again!

[haesleinhuepf]

Ah, I presume this is related to #147 and the related fix. I'll do a little modification in a minute...

[haesleinhuepf]

You know what, I think git-bob can do this.

git-bob think about this!

[github-actions]

^{This message was generated by git-bob (version: 0.2.3, model: gpt-4o-2024-08-06), an experimental AI-based assistant. It can make mistakes and has limitations. Check its messages carefully.}

Hi @thawn,

The error UnboundLocalError: cannot access local variable 'instructions' where it is not associated with a value in generate_code_samples indicates that the instructions variable is being referenced before assignment in certain conditions. To resolve this, make sure the instructions variable is initialized before checking or using it.

Here's a minimal fix in the generate_code_samples function to ensure instructions is initialized:

# Initialize instructions before the try block
instructions = ""
try:
    instructions = func()
except:
    pass

This should prevent the error by ensuring instructions always has a default empty string value. Let me know if the issue persists after applying the fix.

Best,
git-bob

[haesleinhuepf]

Fantastic. git-bob implement exactly this!

[haesleinhuepf]

Hm, let's add the statement before the try: statement.

git-bob solve

[haesleinhuepf]

Ok, git-bob didn't manage, but I released bia-bob 0.21.2 Would you mind checking if it works for you @thawn ?

pip install bia-bob==0.21.2

[thawn]

@haesleinhuepf thanks a lot for the lightning-fast fix! I just tested it on my machine and it works perfectly.

FYI, I had to switch to the model 'alias-large' to get a useful response.

[haesleinhuepf]

Awesome, thanks for the feedback!

I must admit, I haven't used blablador for a while (happy that it still works) and get the most ouf of bia-bob using claude-3.5-sonnet these days. The new gpt4o-2024-08-06 has the longest output (16k tokens) btw in case you're generating really long code snippets ;-)

Thanks again!