#LLM binding is nil

1 messages · Page 1 of 1 (latest)

kind helm
#

I'm a bit confused by this error :
binding value is nil

Here are the logs :

 binding value is nil
✘ Widgetv2.document: Container! 9.2s
! binding value is nil
│🧑             You are an expert at reading code. You have access to a workspace.                                                                                                                                       
│ ┃             Use the default directory in the workspace.                                                                                                                                                              
│ ┃             Your assignment is: explain the folder structure
│ ┃ 0.0s
│ 
│🤖 0.9s
│ ✘ selectToyWorkspace: String! 0.0s
│ ! required argument id not provided
│🤖 0.8s
│ ✘ selectToyWorkspace 0.0s
│ ! arg "id" does not match pattern ^ToyWorkspace#\d+$: "your-workspace-id"
│🤖 0.9s
│ ✘ selectToyWorkspace 0.0s
│ ! arg "id" does not match pattern ^ToyWorkspace#\d+$: "your-workspace-id"
│🤖 1.3s
│ ✘ selectToyWorkspace 0.0s
│ ! arg "id" does not match pattern ^ToyWorkspace#\d+$: "your-workspace-id"
│🤖 1.7s
│ ✘ assignment 0.0s
│ ! unknown arg: "id"
│🤖 1.3s
│ ✘ assignment 0.0s
│ ! unknown arg: "after"
│🤖 0.4s
│ ! model finished: stop
inner garnet
#

@kind helm sorry about that! Are you on Dagger 0.18.3 or 0.18.2?

If you're on 0.18.3, can you try to specify the model explicitly to the LLM in your code. Maybe you're using gpt, or claude, or gemini?

kind helm
#

dagger 0.18.2

#

but i can upgrade !

#

this was on qwen if my memory serves me well

#

any model you recommend i should try this on ?

inner garnet
kind helm
#

trying the new quickstart now ( seems like the code change a bit since yesterday)

#

yeah local models with ollama

#

still getting binding is nil with the quickstart . tried qwen2.5-coder:3b model

#

the formatting is cooked i'm not sure why

#

this is on dagger 0.18.3

inner garnet
#

I'm guessing the 3b is tool small to succeed.

frosty imp
#

I can try to add some more details tomorrow for you, but basically that error means when you selected the output env().output("completed") that completed wasn't assigned which happens when the LLM got lost or confused. In the original output you pasted, you can see it fail a few times trying to figure out tool calling before giving up. With a model as small as 3b, you'll generally need quite a bit of prompting to find success. I'll play around with that model tomorrow and see if I can find some prompts that work for it!

inner garnet
#

Mine got tripped up trying to run a createFile tool, so I amended line 2. Mine also was failing to build, so I amended line 3.

You are an expert Go programmer with an assignment to create a Go program
Create files with WithNewFile tool in the default directory in $builder
Always build the code with 'go build ./...' to make sure it is valid
Do not stop until your assignment is completed and the code builds
Your assignment is: $assignment

was still a bit too hard for my qwen2.5:7B-F16 but was getting closer

kind helm
#

so you recommend using the biggest qwen model ?

inner garnet
#

agree, I'm seeing some garbled TUI output

#

@versed flame 👆