OpenInterpreter
Pin litellm version 1.50.0 while 1.50.2 is getting fixed
Describe the bug
Instead of showing code blocks, it shows
<tool_call>
{"name": "execute", "arguments": {"code": "import os\n# List all .pdf files in the current
directory\npdf_files = [f for f in os.listdir('.') if f.endswith('.pdf')]\npdf_files",
"language": "python"}} </tool_call>
for example
Reproduce
(i1) user@user-PC:~/workspaces/interpreter/i1$ interpreter --disable_telemetry --context_window 11000 --api_base "http://localhost:11434" --no-llm_supports_vision --model ollama_chat/qwen2.5:32b-instruct-q4_K_M_11k_fullgpu --max_output 60000 -y
▌ A new version of Open Interpreter is available.
▌ Please run: pip install --upgrade open-interpreter
───────────────────────────────────────────────────────────────────────────────────────────────────
> cat a part of the pdf, I don't remember what it is
To accomplish this task, we will use a Python library called PyPDF2 to read and extract text
from a PDF file. Since you're not sure which PDF you want to cat (view) a part of, let's start
by listing all the PDF files in your current directory.
First, we'll need to install the PyPDF2 package if it's not already installed:
pip install PyPDF2
Then, we can list out the PDF files for you to choose from. Let's do that now. <tool_call>
{"name": "execute", "arguments": {"code": "import os\n# List all .pdf files in the current
directory\npdf_files = [f for f in os.listdir('.') if f.endswith('.pdf')]\npdf_files",
"language": "python"}} </tool_call>
Expected behavior
(i1) user@user-PC:~/workspaces/interpreter/i1$ uv pip install -U litellm==1.50
Using Python 3.11.10 environment at /home/user/miniconda3/envs/i1
Resolved 45 packages in 195ms
Prepared 1 package in 0.34ms
Uninstalled 1 package in 16ms
Installed 1 package in 14ms
- litellm==1.50.2
+ litellm==1.50.0
(i1) user@user-PC:~/workspaces/interpreter/i1$ interpreter --disable_telemetry --context_window 11000 --api_base "http://localhost:11434" --no-llm_supports_vision --model ollama_chat/qwen2.5:32b-instruct-q4_K_M_11k_fullgpu --max_output 60000 -y
> cat a part of the pdf, I don't remember what it is
To accomplish this task, we need to determine which PDF you're referring to and then extract a
portion of its content.
Let's start by searching for PDF files in your current directory:
import os
# List all .pdf files in the current directory.
pdf_files = [f for f in os.listdir() if f.endswith('.pdf')]
for i, pdf_file in enumerate(pdf_files):
print(f"{i + 1}: {pdf_file}")
if not pdf_files:
print("No PDF files found in the current directory.")
1: TA142.pdf
Screenshots
No response
Open Interpreter version
0.4.0
Python version
3.11.10
Operating System name and version
Ubuntix
Additional context
As seen, downgrading litellm from 1.50.2 to 1.50.0 fixes the issue. I only see the behavior with an ollama instance, and not with groq for example.
I guess it's on litellm side, but it might be to consider to pin the litellm version for now because it makes it unsusable for local ollama users in the current state. That's just a suggestion, I don't know how you deal with such situation usually.
Cheers mates
EDIT: Sorry, the copy paste was from an earlier version of OI, I might have messed up my windows or something. But the behavior of course remains there on 0.4.0
@blakkd Is this still an issue on the latest version (0.4.3) when you run interpreter --local to select the ollama model?
I realize I never tried the interactive option before! That said, I don't know what's going on as I currently can't make it work using it. It correctly calls the API to retrieve the model tags but then it exits with the following error:
Invalid URL '0.0.0.0/api/tags': No scheme supplied. Perhaps you meant https://0.0.0.0/api/tags?
Here are the logs:
litellm
Installed 1 package in 176ms
- litellm==1.50.0
+ litellm==1.51.0
without --local
~ ❯❯❯ interpreter --disable_telemetry --context_window 11000 --api_base "http://localhost:11434" --no-llm_supports_vision --model ollama_chat/qwen2.5:32b-instruct-q4_K_M_11k_fullgpu --max_output 20000 -y
> What time is it in Seattle?
To find out the current time in Seattle, we can use Python's datetime module along with the pytz library to
get the timezone information for Seattle.
Let's start by getting the current time in Seattle. <tool_call> {"name": "execute", "arguments": {"code":
"import datetime\nimport pytz\n\nseattle_tz = pytz.timezone('America/Los_Angeles')\nseattle_time =
datetime.datetime.now(seattle_tz)\nseattle_time.strftime('%Y-%m-%d %H:%M:%S')", "language": "python"}}
</tool_call>
>
Exiting...
with --local
~ ❯❯❯ interpreter --local --disable_telemetry --context_window 11000 --no-llm_supports_vision --max_output 20000 -y
Open Interpreter supports multiple local model providers.
[?] Select a provider:
> Ollama
Llamafile
LM Studio
Jan
[?] Select a model:
deepseek-coder-v2:16b-lite-instruct-q6_K_29k_fullgpu_miro2
deepseek-coder-v2:16b-lite-instruct-q6_K
qwen2.5:32b-instruct-q4_K_M_24k_miro2
qwen2.5:32b-instruct-q4_K_M_11k_fullgpu_miro2
> qwen2.5:32b-instruct-q4_K_M_11k_fullgpu
mistral-nemo:12b-instruct-2407-q8_0_32k
minicpm-v:8b-2.6-fp16
mistral-small:22b-instruct-2409-q6_K_32k
bge-m3_gpu
bge-m3_cpu
nomic-embed-text_cpu
Invalid URL '0.0.0.0/api/tags': No scheme supplied. Perhaps you meant https://0.0.0.0/api/tags?
▌ Ollama not found
Please download Ollama from ollama.com to use qwen2.5:32b-instruct-q4_K_M_11k_fullgpu.
ollama serve
[GIN] 2024/10/28 - 22:05:00 | 200 | 20.799µs | 127.0.0.1 | HEAD "/"
[GIN] 2024/10/28 - 22:05:00 | 200 | 3.081969ms | 127.0.0.1 | GET "/api/tags"