dspy watsonx

[shashanka300]

---

NameError Traceback (most recent call last)
Cell In[20], line 2
1 import dspy
----> 2 lm = dspy.Watsonx(model="mistralai/mixtral-8x7b-instruct-v01", project_id = PROJECT_ID,
3 credentials = CREDENTIALS)
4 dspy.configure(lm=lm)

File ~/Documents/Assets/Dspytest/dsp/lib/python3.11/site-packages/dsp/modules/watsonx.py:65, in Watsonx.init(self, model, credentials, project_id, **kwargs)
53 self.model_type = "instruct"
54 self.kwargs = {
55 "temperature": 0,
56 "decoding_method": "greedy",
(...)
62 **kwargs,
63 }
---> 65 self.client = Model(
66 model_id=self.model,
67 params=self.kwargs,
68 credentials=self.credentials,
69 project_id=self.project_id,
70 )
72 self.history: list[dict[str, Any]] = []

NameError: name 'Model' is not defined

I get this error enven after installing ibm-watsonx-ai

[arnavsinghvi11]

Hi @shashanka300

Can you try running the import locally from the Watson module's import logic? This should be working locally for the module to work.

import ibm_watsonx_ai  # noqa: F401
from ibm_watsonx_ai.foundation_models import Model

Also, I recommend using the latest dspy.LM client over dspy.Watsonx (which will be deprecated in upcoming DSPy releases). dspy.LM uses LiteLLM which supports WatsonX so you can just do something like dspy.LM(model="watsonx/...", other_watsonX_params..... Feel free to check out the migration notebook for more details!

[shashanka300]

hi @arnavsinghvi11
Have tried this before i posted the bug report.
Unfortunately that dosen't work either.

[shashanka300]

currently i am using it with litellm backend

[okhat]

Yes please use dspy.LM