alway respond user instruction
#1
by
telehan
- opened
thanks for your gguf model, but it seems always respond user words
test with ollama with following configuration
$ ollama show --modelfile mistral:7b-ins-fcjson-v0.2-q8
# Modelfile generate by "ollama show"
# To build a new Modelfile based on this, replace FROM with:
# FROM mistral:7b-ins-fcjson-v0.2-q8
FROM ～/.ollama/models/blobs/sha256-b6e8a579e3f7f488d3f9ae7985fc7240a01684f0bfe022534de5774eba0780a6
TEMPLATE [INST] {{ .System }} {{ .Prompt }} [/INST]
PARAMETER num_ctx 4096
PARAMETER stop [INST]
PARAMETER stop [/INST]
$ ollama run mistral:7b-ins-fcjson-v0.2-q8
>>> hi
hi
>>> nice to see u
nice to see u
>>> what's your name
what's your name
hi
@telehan
, thanks for your interest in our model..
You are right, right now for regular chat, this version of model always responds user words.
This model is designed and fine-tuned for function calling task, however it seems that it overfits on function calling data.
We are going to publish new versions that fixes this problem soon.